Example: Factorial design applied in optimisation technique.
To meet the ethical considerations, you need to ensure that.
Collect the data by using suitable data collection according to your experiment’s requirement, such as observations, case studies , surveys , interviews , questionnaires, etc. Analyse the obtained information.
Write the report of your research. Present, conclude, and explain the outcomes of your study .
What is the first step in conducting an experimental research.
The first step in conducting experimental research is to define your research question or hypothesis. Clearly outline the purpose and expectations of your experiment to guide the entire research process.
What are the different types of research you can use in your dissertation? Here are some guidelines to help you choose a research strategy that would make your research more credible.
Descriptive research is carried out to describe current issues, programs, and provides information about the issue through surveys and various fact-finding methods.
A hypothesis is a research question that has to be proved correct or incorrect through hypothesis testing – a scientific approach to test a hypothesis.
USEFUL LINKS
LEARNING RESOURCES
COMPANY DETAILS
Saul McLeod, PhD
Editor-in-Chief for Simply Psychology
BSc (Hons) Psychology, MRes, PhD, University of Manchester
Saul McLeod, PhD., is a qualified psychology teacher with over 18 years of experience in further and higher education. He has been published in peer-reviewed journals, including the Journal of Clinical Psychology.
Learn about our Editorial Process
Olivia Guy-Evans, MSc
Associate Editor for Simply Psychology
BSc (Hons) Psychology, MSc Psychology of Education
Olivia Guy-Evans is a writer and associate editor for Simply Psychology. She has previously worked in healthcare and educational sectors.
On This Page:
The experimental method involves the manipulation of variables to establish cause-and-effect relationships. The key features are controlled methods and the random allocation of participants into controlled and experimental groups .
An experiment is an investigation in which a hypothesis is scientifically tested. An independent variable (the cause) is manipulated in an experiment, and the dependent variable (the effect) is measured; any extraneous variables are controlled.
An advantage is that experiments should be objective. The researcher’s views and opinions should not affect a study’s results. This is good as it makes the data more valid and less biased.
There are three types of experiments you need to know:
A laboratory experiment in psychology is a research method in which the experimenter manipulates one or more independent variables and measures the effects on the dependent variable under controlled conditions.
A laboratory experiment is conducted under highly controlled conditions (not necessarily a laboratory) where accurate measurements are possible.
The researcher uses a standardized procedure to determine where the experiment will take place, at what time, with which participants, and in what circumstances.
Participants are randomly allocated to each independent variable group.
Examples are Milgram’s experiment on obedience and Loftus and Palmer’s car crash study .
A field experiment is a research method in psychology that takes place in a natural, real-world setting. It is similar to a laboratory experiment in that the experimenter manipulates one or more independent variables and measures the effects on the dependent variable.
However, in a field experiment, the participants are unaware they are being studied, and the experimenter has less control over the extraneous variables .
Field experiments are often used to study social phenomena, such as altruism, obedience, and persuasion. They are also used to test the effectiveness of interventions in real-world settings, such as educational programs and public health campaigns.
An example is Holfing’s hospital study on obedience .
A natural experiment in psychology is a research method in which the experimenter observes the effects of a naturally occurring event or situation on the dependent variable without manipulating any variables.
Natural experiments are conducted in the day (i.e., real life) environment of the participants, but here, the experimenter has no control over the independent variable as it occurs naturally in real life.
Natural experiments are often used to study psychological phenomena that would be difficult or unethical to study in a laboratory setting, such as the effects of natural disasters, policy changes, or social movements.
For example, Hodges and Tizard’s attachment research (1989) compared the long-term development of children who have been adopted, fostered, or returned to their mothers with a control group of children who had spent all their lives in their biological families.
Here is a fictional example of a natural experiment in psychology:
Researchers might compare academic achievement rates among students born before and after a major policy change that increased funding for education.
In this case, the independent variable is the timing of the policy change, and the dependent variable is academic achievement. The researchers would not be able to manipulate the independent variable, but they could observe its effects on the dependent variable.
Ecological validity.
The degree to which an investigation represents real-life experiences.
These are the ways that the experimenter can accidentally influence the participant through their appearance or behavior.
The clues in an experiment lead the participants to think they know what the researcher is looking for (e.g., the experimenter’s body language).
The variable the experimenter manipulates (i.e., changes) is assumed to have a direct effect on the dependent variable.
Variable the experimenter measures. This is the outcome (i.e., the result) of a study.
All variables which are not independent variables but could affect the results (DV) of the experiment. EVs should be controlled where possible.
Variable(s) that have affected the results (DV), apart from the IV. A confounding variable could be an extraneous variable that has not been controlled.
Randomly allocating participants to independent variable conditions means that all participants should have an equal chance of participating in each condition.
The principle of random allocation is to avoid bias in how the experiment is carried out and limit the effects of participant variables.
Changes in participants’ performance due to their repeating the same or similar test more than once. Examples of order effects include:
(i) practice effect: an improvement in performance on a task due to repetition, for example, because of familiarity with the task;
(ii) fatigue effect: a decrease in performance of a task due to repetition, for example, because of boredom or tiredness.
We provide new evidence on the causal effect of unearned income on consumption, balance sheets, and financial outcomes by exploiting an experiment that randomly assigned 1000 individuals to receive $1000 per month and 2000 individuals to receive $50 per month for three years. The transfer increased measured household expenditures by at least $300 per month. The spending impact is positive in most categories, and is largest for housing, food, and car expenses. The treatment increases housing unit and neighborhood mobility. We find noisily estimated modest positive effects on asset values, driven by financial assets, but these gains are offset by higher debt, resulting in a near-zero effect on net worth. The transfer increased self-reported financial health and credit scores but did not affect credit limits, delinquencies, utilization, bankruptcies, or foreclosures. Adjusting for underreporting, we estimate marginal propensities to consume non-durables between 0.44 and 0.55, durables and semi-durables between 0.21 and 0.26, and marginal propensities to de-lever of near zero. These results suggest that large temporary transfers increase short-term consumption and improve financial health but may not cause persistent improvements in the financial position of young, low-income households.
Many people contributed to the success of this project. The program we study and the associated research were supported by generous private funding sources, and we thank the non-profit organizations that implemented the program. We thank Jill Adona, Isaac Ahuvia, Oscar Alonso, Francisco Brady, Jack Bunge, Jake Cosgrove, Leo Dai, Kevin Didi, Rashad Dixon, Marc-Andrea Fiorina, Joshua Lin, Sabrina Liu, Anthony McCanny, Janna Mangasep, Oliver Scott Pankratz, Alok Ranjan, Mark Rick, Ethan Sansom, Sophia Scaglioni, and Angela Wang-Lin for outstanding research assistance. Tess Cotter, Karina Dotson, Aristia Kinis, Sam Manning, Alex Nawar, and Elizabeth Proehl were invaluable contributors through their work at OpenResearch. The management and staff of the Inclusive Economy Lab at the University of Chicago, including Carmelo Barbaro, Janelle Blackwood, Katie Buitrago, Melinda Croes, Crystal Godina, Kelly Hallberg, Kirsten Jacobson, Timi Koyejo, Misuzu Schexnider, Stephen Stapleton, and many others have provided important support throughout all stages of the project. We received valuable feedback on the study from the OpenResearch Advisory Board and seminar participants at the University of California-Berkeley and the University of Illinois at Urbana-Champaign. This study was approved by the Advarra Institutional Review Board (IRB) and is pre-registered at the American Economic Association RCT registry with a registration ID of AEARCTR-0006750. This research was supported in part by a J-PAL grant titled "The Impact of Unconditional Cash Transfers on Consumption: Evidence from the United States." The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research.
MARC RIS BibTeΧ
Download Citation Data
More from nber.
In addition to working papers , the NBER disseminates affiliates’ latest findings through a range of free periodicals — the NBER Reporter , the NBER Digest , the Bulletin on Retirement and Disability , the Bulletin on Health , and the Bulletin on Entrepreneurship — as well as online conference reports , video lectures , and interviews .
You are accessing a machine-readable page. In order to be human-readable, please install an RSS reader.
All articles published by MDPI are made immediately available worldwide under an open access license. No special permission is required to reuse all or part of the article published by MDPI, including figures and tables. For articles published under an open access Creative Common CC BY license, any part of the article may be reused without permission provided that the original article is clearly cited. For more information, please refer to https://www.mdpi.com/openaccess .
Feature papers represent the most advanced research with significant potential for high impact in the field. A Feature Paper should be a substantial original Article that involves several techniques or approaches, provides an outlook for future research directions and describes possible research applications.
Feature papers are submitted upon individual invitation or recommendation by the scientific editors and must receive positive feedback from the reviewers.
Editor’s Choice articles are based on recommendations by the scientific editors of MDPI journals from around the world. Editors select a small number of articles recently published in the journal that they believe will be particularly interesting to readers, or important in the respective research area. The aim is to provide a snapshot of some of the most exciting work published in the various research areas of the journal.
Original Submission Date Received: .
Find support for a specific problem in the support section of our website.
Please let us know what you think of our products and services.
Visit our dedicated information section to learn more about MDPI.
Experimental research on the low-cycle fatigue crack growth rate for a stiffened plate of eh36 steel for use in ship structures.
2. low cycle fatigue crack growth experiment for stiffened plate, 3. result and discussion, 3.1. experimental results of stiffened plates with single-edge crack, 3.2. experimental results of stiffened plates with central crack, 4. conclusions, author contributions, institutional review board statement, informed consent statement, data availability statement, conflicts of interest.
Click here to enlarge figure
Elastic Modulus/GPa | Poisson’s Ratio | Yield Stress/MPa | Ultimate Tensile Strength/MPa |
---|---|---|---|
206 | 0.3 | 434.94 | 548.91 |
Specimen Number | P /kN | R = P /P | Nominal Stress/MPa | Crack Location | Stiffener Height |
---|---|---|---|---|---|
P1 | 84.24 | −1 | 120 | single-edge crack | 30 mm |
P2 | 90.72 | −1 | 130 | single-edge crack | 30 mm |
P3 | 97.20 | −1 | 140 | single-edge crack | 30 mm |
P4 | 384.00 | 0.031 | 280 | central crack | 30 mm |
P5 | 420.00 | 0.2 | 300 | central crack | 30 mm |
P6 | 420.00 | 0.2 | 300 | central crack | 0 mm |
The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
Dong, Q.; Xu, G.; Chen, W. Experimental Research on the Low-Cycle Fatigue Crack Growth Rate for a Stiffened Plate of EH36 Steel for Use in Ship Structures. J. Mar. Sci. Eng. 2024 , 12 , 1365. https://doi.org/10.3390/jmse12081365
Dong Q, Xu G, Chen W. Experimental Research on the Low-Cycle Fatigue Crack Growth Rate for a Stiffened Plate of EH36 Steel for Use in Ship Structures. Journal of Marine Science and Engineering . 2024; 12(8):1365. https://doi.org/10.3390/jmse12081365
Dong, Qin, Geng Xu, and Wei Chen. 2024. "Experimental Research on the Low-Cycle Fatigue Crack Growth Rate for a Stiffened Plate of EH36 Steel for Use in Ship Structures" Journal of Marine Science and Engineering 12, no. 8: 1365. https://doi.org/10.3390/jmse12081365
Article access statistics, further information, mdpi initiatives, follow mdpi.
Subscribe to receive issue release notifications and newsletters from MDPI journals
Office of Science Policy
NIH promotes the safe and responsible use of AI in biomedical research through programs that support the development and use of algorithms and models for research, contribute to AI-ready datasets that accelerate discovery, and encourage multi-disciplinary partnerships that drive transparency, privacy, and equity.
Advancements in artificial intelligence (AI) are spurring tremendous progress in medical research to enhance human health and longevity. To that end, NIH has a robust system of policies and practices that guide stakeholders across the biomedical and behavioral research ecosystem. While AI may not be explicitly mentioned, NIH’s policy framework is designed to responsibly guide and govern advancing science and emerging technologies, including development and use of AI technologies in research.
The policies, best practices, and regulations listed below reflect this framework and should be considered before, during, and after development and use of AI in research. This is not an exhaustive list of all policies and requirements that may apply to any NIH-supported research project but can serve as a guide for the research community.
Please note: Unauthorized data disclosures violate several of the policies listed below. Investigators should be cognizant that research data used as input or training for AI could result in their unintentional disclosure if the data is sent to an AI provider external to NIH.
The following establish expectations and best practices for protecting the welfare, privacy, and autonomy of research participants. The ethical considerations embedded in these policies, regulations, and best practices (e.g., privacy) address key issues relevant to the development and use of AI in research. In adhering to them, investigators can mitigate potential harms and inequities arising from the use and development of AI.
Protection of Human Subjects (45 CFR 46) : Outlines basic provisions for Institutional Review Boards, informed consent, and assurance of compliance for NIH-supported research involving human participants and their data, including considerations of risks & benefits.
For clinical investigations that are also regulated by the Food and Drug Administration, see:
21 CFR 50 Protection of Human Subjects 21 CFR 56 Institutional Review Boards
Certificates of Confidentiality : Prohibits the disclosure of identifiable, sensitive research information to anyone not connected to the research except when the participant consents or in a few other specific situations.
NIH Information about Protecting Privacy When Sharing Human Research Participant Data : Provides a set of principles and best practices for protecting the privacy of human research participants when sharing data in NIH-supported research. (Issued under the NIH Data Management and Sharing policy.)
NIH Informed Consent for Secondary Research with Data and Biospecimens : Provides points to consider, instructions for use, and optional sample language that is designed for informed consent documents for research studies that include plans to store and share collected data and biospecimens for future use.
The following seek to maximize the responsible management and sharing of scientific data while ensuring that researchers consider how the privacy, rights, and confidentiality of human research participants will be protected. Increasing the availability of data through data sharing allows for more accurate development and use of AI models. These policies help ensure that investigators remain good stewards of data used in or produced by AI models.
NIH Data Management & Sharing (DMS) Policy : Establishes the requirement to submit a DMS Plan and comply with NIH-approved plans. In addition, NIH Institutes, Centers, and Offices can request additional or specific information be included within the plan to support programmatic priorities or to expand the utility of the scientific data generated from the research. Also see DMS Policy Frequently Asked Questions .
Responsible Management and Sharing of American Indian/Alaska Native (AI/AN) Participant Data : Describes considerations and best practices for the responsible and respectful management and sharing of AI/AN participant data under the DMS Policy.
NIH Genomic Data Sharing Policy : Promotes and facilitates responsible sharing of large-scale genomic data generated with NIH funds. Also see Genomic Data Sharing Frequently Asked Questions .
Health Insurance Portability and Accountability Act (HIPAA) helps protect the privacy and security of health data used in research, including research involving AI, thereby fostering trust in healthcare research activities.
HIPAA Privacy Rule : Establishes the conditions under which protected health information may be used or disclosed by covered entities for research purposes.
The following establish guidance, expectations, and best practices related to intellectual property and software sharing. They complement NIH’s data sharing initiatives, delineate investigator rights under the SBIR and STTR programs, and provide USPTO guidance on AI-related inventions. While many are not specific to AI, the policies and programs below are relevant to investigators who have developed software and source code under NIH research grants or who intend to commercialize their NIH-supported research products, including those related to development and use of AI.
NIH Best Practices for Sharing Research Software : Best practices for sharing research software and source code in a free and open format.
NIH Small Business Innovation Research (SBIR) & Small Business Technology Transfer (STTR) : Unique policies and approaches may apply in the context of NIH’s Small Business Innovation Research (SBIR) & Small Business Technology Transfer (STTR) program. For example, recipients may retain the rights to data generated during the performance of an SBIR or STTR award.
NIH Research Tools Policy : NIH expects funding recipients to appropriately disseminate propagate and allow open access to research tools developed with NIH funding.
US Patent and Trademark Office information about AI : Provides AI-related patent resources and important information concerning AI IP policy.
The following clarifies NIH’s stance on the use of generative AI tools during peer review.
NOT-OD-23-149: Informs the extramural community that the NIH prohibits NIH scientific peer reviewers from using natural language processors, large language models, or other generative AI technologies for analyzing and formulating peer review critiques for grant applications and R&D contract proposals. Also see Open Mike blog on Using AI in Peer Review Is a Breach of Confidentiality .
The following establish and are part of a comprehensive biosecurity and biosafety oversight system. Research funded by NIH, including research using the tools and technologies enabled or informed by AI, fall under this oversight framework. While some of these policies do not explicitly address AI, they are still applicable to development and use of AI in research involving biological agents, toxins, or nucleic acid molecules if such research involves physical experiments that are covered under these policies.
United States Government Policy for Oversight of Life Sciences Dual Use Research of Concern : Describes practices and procedures to ensure that dual use research of concern (DURC) is identified at the institutional level and risk mitigation measures are implemented as necessary for U.S. Government-funded research. DURC is “life sciences research that, based on current understanding, can be reasonably anticipated to provide knowledge, information, products, or technologies that could be directly misapplied to pose a significant threat with broad potential consequences to public health and safety, agricultural crops and other plants, animals, the environment, materiel, or national security.” The United States Government Policy for Institutional Oversight of Life Sciences Dual Use Research of Concern complements the aforementioned policy and addresses institutional oversight of DURC, which includes policies, practices, and procedures to ensure DURC is identified and risk mitigation measures are implemented, where applicable.
HHS Framework for Guiding Funding Decisions about Proposed Research Involving Enhanced Potential Pandemic Pathogens (HHS P3CO Framework): Guides Department of Health and Human Services funding decisions on individual proposed research that is reasonably anticipated to create, transfer, or use enhanced potential pandemic pathogens (ePPP). ePPP research is research that “may be reasonably anticipated to create, transfer or use potential pandemic pathogens resulting from the enhancement of a pathogen’s transmissibility and/or virulence in humans.” The HHS P3CO Framework is responsive to and in accordance with the Recommended Policy Guidance for Departmental Development of Review Mechanisms for Potential Pandemic Pathogen Care and Oversight issued in 2017 by the White House Office of Science and Technology Policy.
United States Government Policy for Oversight of Dual Use Research of Concern and Pathogens with Enhanced Pandemic Potential : On May 6, 2024, the White House Office of Science and Technology Policy released this new policy along with associated Implementation Guidance . This will supersede the DURC and P3CO policy frameworks on May 6, 2025. It provides a unified federal oversight framework for conducting and managing certain types of federally funded life sciences research on biological agents and toxins that have the potential to pose risks to public health, agriculture, food security, economic security, or national security. The policy “encourages institutional oversight of in silico research, regardless of funding source, that could result in the development of potential dual-use computational models directly enabling the design of a [pathogen with enhanced pandemic potential] or a novel biological agent or toxin.”
NIH Guidelines for Research Involving Recombinant or Synthetic Nucleic Acid Molecules : Establish safety practices and containment procedures for institutions that receive NIH funding for “basic and clinical research involving recombinant or synthetic nucleic acid molecules, including the creation and use of organisms and viruses containing recombinant or synthetic nucleic acid molecules.”
For regulatory questions related to AI, see:
An official website of the United States government
Here’s how you know
Official websites use .gov A .gov website belongs to an official government organization in the United States.
Secure .gov websites use HTTPS A lock ( Lock A locked padlock ) or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.
JavaScript appears to be disabled on this computer. Please click here to see any active alerts .
Published August 7, 2024
To some people, tire pollution might draw up an image of a blown-out or discarded tire on the side of a highway, or stockpiled old tires behind a garage. However, the issue of tire pollution is more complex and prolific than at first glance, as every step of a tire’s life cycle, from production to use to disposal, can impact our environment, health and wildlife.
Meet EPA Ecologist Paul Mayer, Ph.D.
Meet Our Other Ecosystem Researchers
To address growing concerns of tire pollution and a specific pollutant called 6PPD-quinone (6PPD-Q) , EPA researcher Dr. Paul Mayer led an effort to investigate the life cycle of tires and their impacts on the environment. The resulting article, “ Where the rubber meets the road: Emerging environmental impacts of tire wear particles and their chemical cocktails ,” is a holistic examination and data compilation of tires as complex pollutants across three levels: their whole state (e.g., tire production or disposal in landfills), as particulates (i.e., as they are worn down), and as “chemical cocktails.”
The research team illustrated that the production of over 3 billion tires annually requires massive amounts of natural resources, including fossil fuels, water, and agricultural space to grow natural rubber, which has been linked to deforestation. The manufacturing process involves chemical mixtures that emit carcinogens (cancer-causing substances) and radioactive compounds. Over 800 million tires are disposed of annually and burned for fuel or broken down and recycled into products such as artificial turf infill, asphalt, landscape mulch and doormats. These processes may introduce hazards such as contact exposure to chemicals and heavy metals, inhalation, ingestion, and other risks associated with tire crumb. Further, tire piles can catch fire and burn for long periods of time, emitting harmful pollutants such as fine particulate matter (PM2.5) .
The researchers found that one tire will shed between two and fourteen pounds of rubber particles due to road wear (from initial use to initial disposal). These particles may be small enough to be picked up by wind and carried for up to a month before they are deposited on land. Larger particles can be caught in stormwater runoff and transported along curbs and through stormwater systems where they are typically deposited into a local waterway. Constituents of these particles, pollutants such as microplastics , heavy metals, hydrocarbons, and other toxic chemicals can then pollute local water and soil.
The researchers also conducted a life cycle analysis of rubber tires, following one product unit from creation to disposal, identifying information gaps in tire related research along the way. The rate and volume of tire wear particle release may differ between tire brands and types. The size, shape, and surface properties of tire particles can impact the methods of their emission and transport. Further research is also needed to characterize the toxicity of tire pollutants and their health effects, including determining alternative chemicals for use in the manufacturing process and conducting longer term studies on populations of sensitive species. More accurate data on tire particle and chemical emissions based on climate, population density, and transportation infrastructure is needed to support the development of effective methods of tire pollution reduction, remediation, and risk management. These information gaps and many others identified by the research team show that tire wear particles and chemicals present a strong risk to human health and the environment, and action should be taken to research and mitigate this issue.
Several research teams across the EPA are working on addressing information gaps specifically related to the pollutant 6PPD-Q. 6PPD-Q is the product of a reaction between 6PPD, a chemical added in the tire manufacturing process, and ozone in the air. EPA-funded research in 2020 showed 6PPD-Q in stormwater to be highly toxic to several salmonid fish species and lethal to the threatened and endangered populations of coho salmon. This species is a culturally, economically, and ecologically important resource for many Tribal nations along the Pacific Northwest coast and its connected waterways. Healthy and accessible salmon populations are critical to the health and wellbeing of Tribes, including the practice and protection of Tribal Treaty Rights.
EPA ecologist Dr. Jonathan Halama is using the advanced EPA model Visualizing Ecosystem Land Management Assessments (VELMA) to learn more about the fate and transport of 6PPD-Q from tire particles in stormwater. Through the analysis of current stormwater management systems and estimated roadway deposition patterns based on traffic count data, Halama and his team are working to understand the processes influencing tire particle flow paths and to determine hotspots where 6PPD-Q is concentrated within a watershed. Using VELMA to find these 6PPD-Q hotspots can help researchers prioritize the locations and types of stormwater management designs to reduce 6PPD-Q levels most effectively.
In 2023, the EPA developed a draft analytical method to identify 6PPD-Q in surface waters and stormwater. In addition to tire life cycle analysis and stormwater management modeling, there are multiple research efforts within the EPA and in collaboration with external partners that focus on 6PPD-Q. EPA researchers are developing measurement methods for 6PPD-Q in air and sediment, tools to screen the toxicity of environmental samples, and health hazard screening values. To further protect coho salmon and other sensitive aquatic species, researchers are also investigating brake and tire emission rates of particulates, 6PPD, and metals, health effects of tire wear particles and 6PPD-Q on aquatic life, and potential alternative chemicals to 6PPD in tires.
Dr. Mayer presented about tires as complex pollutants at EPA’s Water Research Webinar on June 26th, 2024. You can watch a recording of the session here .
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
npj Unconventional Computing volume 1 , Article number: 3 ( 2024 ) Cite this article
4790 Accesses
240 Altmetric
Metrics details
The conventional computing paradigm struggles to fulfill the rapidly growing demands from emerging applications, especially those for machine intelligence because much of the power and energy is consumed by constant data transfers between logic and memory modules. A new paradigm, called “computational random-access memory (CRAM),” has emerged to address this fundamental limitation. CRAM performs logic operations directly using the memory cells themselves, without having the data ever leave the memory. The energy and performance benefits of CRAM for both conventional and emerging applications have been well established by prior numerical studies. However, there is a lack of experimental demonstration and study of CRAM to evaluate its computational accuracy, which is a realistic and application-critical metric for its technological feasibility and competitiveness. In this work, a CRAM array based on magnetic tunnel junctions (MTJs) is experimentally demonstrated. First, basic memory operations, as well as 2-, 3-, and 5-input logic operations, are studied. Then, a 1-bit full adder with two different designs is demonstrated. Based on the experimental results, a suite of models has been developed to characterize the accuracy of CRAM computation. Scalar addition, multiplication, and matrix multiplication, which are essential building blocks for many conventional and machine intelligence applications, are evaluated and show promising accuracy performance. With the confirmation of MTJ-based CRAM’s accuracy, there is a strong case that this technology will have a significant impact on power- and energy-demanding applications of machine intelligence.
Introduction.
Recent advances in machine intelligence 1 , 2 for tasks such as recommender systems 3 , speech recognition 4 , natural language processing 5 , and computer vision 6 , have been placing growing demands on our computing systems, especially for implementations with artificial neural networks. A variety of platforms are used, from general-purpose CPUs and GPUs 7 , 8 , to FPGAs 9 , to custom-designed accelerators and processors 10 , 11 , 12 , 13 , to mixed- or fully- analog circuits 14 , 15 , 16 , 17 , 18 , 19 , 20 . Most are based on the Von Neumann architecture, with separate logic and memory systems. As shown in Fig. 1a , the inherent segregation of logic and memory requires large amounts of data to be transferred between these modules. In data-intensive scenarios, this transfer becomes a major bottleneck in terms of performance, energy consumption, and cost 21 , 22 , 23 . For example, the data movement consumes about 200 times the energy used for computation when reading three 64-bit source operands from and writing one 64-bit destination operand to an off-chip main memory 21 . This bottleneck has long been studied. Research aiming at connecting logic and memory more closely has led to new computation paradigms.
a , b Compared to a conventional computer architecture ( a ), which suffers from the memory-logic transfer bottleneck, CRAM ( b ) offers significant power and performance improvements. Its unique architecture allows for computation in memory, as well as, random access, reconfigurability, and parallel operation capability. c The CRAM could excel in data-intensive, memory-centric, or power-sensitive applications, such as neural networks, image processing, or edge computing ( c ).
Promising paradigms include “near-memory” and “in-memory” computing. Near-memory processing brings logic physically closer to memory by employing 3D-stacked architectures 24 , 25 , 26 , 27 , 28 , 29 . In-memory computing scatters clusters of logic throughout or around the memory banks on a single chip 14 , 15 , 16 , 17 , 18 , 19 , 20 , 30 , 31 , 32 , 33 , 34 , 35 . Yet another approach is to build systems where the memory itself can perform computation. This has been dubbed “true” in-memory computing 36 , 37 , 38 , 39 , 40 , 41 , 42 . The computational random-access memory (CRAM) 38 , 40 is one of the true in-memory computing paradigms. Logic is performed natively by the memory cells; the data for logic operations never has to leave the memory (Fig. 1b ). Additionally, CRAM operates in a fully digital fashion, unlike most other reported in-memory computing schemes 14 , 15 , 16 , 17 , 18 , 19 , 20 , which are partially or mostly analog. CRAM promises superior energy efficiency and processing performance for machine intelligence applications. It has unique additional features, such as random-access of data and operands, massive parallel computing capabilities, and reconfigurability of operations 38 , 40 . Also note that although the transistor-less (crossbar) architecture employed by most of the previous true-in-memory computing paradigms 36 , 37 , 39 , 42 allows for higher density, the maximum allowable size of the memory array is often severely limited due to the sneak path issues. CRAM includes transistors in each of its cells for better-controlled electrical accessibility and, therefore, a larger array size.
The CRAM was initially proposed based on the MTJ device 38 , an emerging memory device that relies on spin electronics 43 . Such “spintronic” devices, along with other non-volatile emerging memory devices, usually referred to as “X” for logic applications, have been intensively investigated over the past several decades for emerging memory and computing applications as “beyond-CMOS” and/or “CMOS + X” technologies. They offer vastly improved speed, energy efficiency, area, and cost. An additional feature that is exploited by CRAM is their non-volatility 44 . The MTJ device is the most mature of spintronic devices for embedded memory applications, based on endurance 45 , energy efficiency 46 , and speed 47 . We note that CRAM can be implemented based not only on spintronics devices but also on other non-volatile emerging memory devices.
In its simplest form, an MTJ consists of a thin tunneling barrier layer sandwiched between two ferromagnetic (FM) layers. When a voltage is applied between the two layers, electrons tunnel through the barrier, resulting in a charge current. The resistance of the MTJ is a function of the magnetic state of the two FM layers, due to the tunneling magnetoresistance (TMR) effect 48 , 49 , 50 . An MTJ can be engineered to be magnetically bi-stable. Accordingly, it can store information based on its magnetic state. This information can be retrieved by reading the resistance of the device. The MTJ can be electrically switched from one state to the other with a current due to the spin-transfer torque (STT) effect 51 , 52 . In this way, an MTJ can be used as an electrically operated memory device with both read and write functionality. A type of random-access memory, the STT-MRAM 53 , 54 , 55 , 56 has been developed commercially, utilizing MTJs as memory cells. A typical STT-MRAM consists of an array of bit cells, each containing one transistor and one MTJ. These are referred to as 1 transistor 1 MTJ (1T1M) cells.
A typical CRAM cell design, as shown in Fig. 2a , is a modification of the 1T1M STT-MRAM architecture 57 . The MTJ, one of the transistors, word line (WL), bit select line (BSL), and memory bit line (MBL) resemble the 1T1M cell architecture of STT-MRAM, which allows the CRAM to perform memory operations. In order to enable logic operations, a second transistor, as well as a logic line (LL) and a logic bit line (LBL), are added to each memory cell. During a logic operation, corresponding transistors and lines are manipulated so that several MTJs in a row are temporarily connected to a shared LL 40 . While the LL is left floating, voltage pulses are applied to the lines connecting to input MTJs, with that of the output MTJ being grounded. The logic operation is based on a working principle called voltage-controlled logic (VCL) 58 , 59 , which utilizes the thresholding effect that occurs when switching an MTJ and the TMR effect of MTJ. As shown in Fig. 2b , when a voltage is applied across the input MTJs, the different resistance values result in different current levels. The current flows through the output MTJ, which may or may not switch its state, depending on the states of the input MTJs. In this way, basic bitwise logic operations, such as AND, OR, NAND, NOR, and MAJ, can be realized. A unique feature of VCL is that the logic operation itself does not require the data in the input MTJs to be read-out through sense amplifiers at the edge of the array. Rather, it is used locally within the set of MTJs involved in the computation. This is fundamentally why the CRAM computation represents true-in-memory computing: the computation does not require data to travel out of the memory array. It is always processed locally by nearby cells. We note that this concept would also work with other two-terminal stateful passive memory devices, such as memristors. Accordingly, a CRAM could be implemented with such devices. A CRAM could also be implemented with three-terminal stateful devices, such as spin-orbit torque (SOT) devices. This could result in greater energy efficiency and reliability 60 . Although devices with progressive or accumulative switching behavior, such as spintronic domain wall devices 61 , 62 , can be adopted as well, CRAM would otherwise work best if adopting bi-stable memory devices with strong threshold switching behavior. As an oversimplified speculation, the performance comparison between CRAMs implemented by various emerging memory devices is expected to roughly follow the comparison between these for memory applications, since CRAM utilizes memory devices in similar manners as in-memory applications. For example, a CRAM implemented based on MTJs should be expected to offer high endurance and high speed. Also, generally, a CRAM logic operation should consume energy comparable to the energy consumption of a memory write operation, for the same emerging memory device operating at the same speed. However, a careful case-by-case analysis is necessary for CRAMs implemented by each emerging memory device technology. Also, note that we do not show a specific circuit design of CRAM peripherals because CRAM does not require significant circuit design change in sense amplifiers or peripherals compared to 1T1M STT-MRAM. And these in the STT-MRAM are already common and mature. Lastly, the true in-memory computing characteristic of CRAM is limited to within a continuous CRAM array: any computation that requires access to data across separate CRAM arrays will require additional data access and movement. The size of an array is ultimately limited by parasitic effects of interconnects 63 . However, these limitations are true for all other in-memory computing paradigms. CRAM is not at any disadvantage in this scenario.
a CRAM adopts the so-called 2 transistor 1 MTJ (2T1M) cell architecture. On top of the 1T1M cell architecture of STT-MRAM, an additional transistor, as well as the added logic line (LL) and logic bit line (LBL), allow the CRAM to perform logic operations. During a CRAM logic operation, the transistors and lines are manipulated to form an equivalent circuit, as shown in b . Although CRAM can be built based on various emerging memory devices, we use MTJs and MTJ-based CRAM as an example for illustration purposes. b The working principle of CRAM logic operation, the VCL, utilizes the thresholding effect that occurs when switching an MTJ and the TMR effect of the MTJ. With an appropriate V logic amplitude, the voltage is translated into different currents flowing through the output MTJ by the TMR effect of the input MTJs. Whether the output MTJ switches or not is dependent on the state of the input MTJs.
On top of the potential performance benefits that the emerging memory devices bring, at circuit and architecture level, CRAM fundamentally provides several benefits (Fig. 1b ): (1) the elimination of the costly performance and energy penalties associated with transferring data between logic and memory; (2) random access of data for the inputs and outputs to operations; (3) the reconfigurability of operations, as any of the logic operations, AND, OR, NAND, NOR, and MAJ can be programmed; and (4) the performance gain of massive parallelism, as identical operations can be performed in parallel in each row of the CRAM array when data is allocated properly. Based on analysis and benchmarking, CRAM has the potential to deliver significant gains in performance and power efficiency, particularly for data-intensive, memory-centric, or power-sensitive applications, such as bioinformatics 40 , 64 , 65 , image 66 and signal 67 processing, neural networks 66 , 68 , and edge computing 69 (Fig. 1c ). For example, a CRAM-based machine-learning inference accelerator was estimated to achieve an improvement on the order of 1000× over a state-of-art solution, in terms of the energy-delay product 70 . Another example shows that CRAM (at the 10 nm technology node) consumes 0.47 µJ and 434 ns of energy and time, respectively, to perform an MNIST handwritten digit classifier task. It is 2500× and 1700× less in energy and time, respectively, compared to a near-memory processing system at the 16 nm technology node 66 . And yet, to date, there have been no experimental studies of CRAM.
In this work, we present the first experimental demonstration of a CRAM array. Although based on a small 1 × 7 array, it successfully shows complete CRAM array operations. We illustrate computation with a 1-bit full adder. This work provides a proof-of-concept as well as a platform with which to study key aspects of the technology experimentally. We provide detailed projections and guidelines for future CRAM design and development. Specifically, based on the experimental results, models and calculations of CRAM logic operations are developed and verified. The results connect the CRAM gate-level accuracy or error rate to MTJ TMR ratio, logic operation pulse width, and other parameters. Then we evaluate the accuracy of a multi-bit adder, a multiplier, and a matrix multiplication unit, which are essential building blocks for many conventional and machine intelligence applications, including artificial neural networks.
Figure 3 shows the experimental setup, consisting of both hardware and software. The hardware is built with a so-called ‘circuit-around-die’ approach 71 : semiconductor circuitry is built with commercially available components around the MTJ dies. This approach offers a more rapid development cycle and flexibility needed for exploratory experimental studies on CRAM arrays and potential new MTJ technologies, while the major foundries lack the specific process design kit available for making a CRAM array fully integrated with CMOS. The hardware is a 1 × 7 CRAM array, with the design of cells taken from the 2T1M CRAM cells 38 , 40 , modified for simplified memory access. Software on a PC controls the operation. It communicates with the hardware with basic commands: ‘open/close transistors’; ‘apply voltage pulses’ to perform write and logic operations; and ‘read cell resistance’. The software collects real-time measurements of the data associated with CRAM operations for analysis and visualization. All aspects of the 1 × 7 CRAM array are functional: memory write, memory read, and logic operations (more details in Methods section, and Supplementary Note S 1 ).
The setup consists of custom-built hardware and a suite of control software. It demonstrates a fully functioning 1 × 7 CRAM array. The hardware consists of a main board hosting all necessary electronics except for the MTJ devices; a connection board on which passive switches, connectors, and magnetic bias field mechanisms are hosted; and multiple cartridge boards that each have an MTJ array mounted and multiple MTJ devices that are wire bonded. The gray-scale scanning electron microscopy image shows the MTJ array used. The color optical photographs show the cartridge board and the entire hardware setup. The software is responsible for real-time measurements of the MTJs; configuration and execution of CRAM operations: memory write, memory read, and logic; and data collection. It is run on a PC, which communicates wirelessly with the main board.
MTJs with perpendicular interfacial anisotropy are used in the CRAM. They exhibit low resistance-area (RA) product and high TMR ratio—approximately 100%—when sized at 100 nm in diameter (more details in Supplementary Note S 2 ).
The experiments begin with measuring the resistance (R)–voltage (V) properties of each MTJ device and of each die. In order to compensate for device-to-device variations, the bias magnetic fields for each MTJ are adjusted so that the R–V properties are as close to each other as possible (more details in Supplementary Note S 2 ). As the processes of making CRAM arrays mature, bias magnetic fields are expected to be no longer needed and all CRAM cells will be able to be operated with uniform parameters and under uniform conditions. The resistance threshold for the MTJs logic states is also determined in this stage.
Then the seven MTJ cells are tested for memory operations with various write pulse amplitudes and widths. Based on the observed write error rates for memory write operations, appropriate pulse amplitudes and widths are configured, achieving reliable memory write operations with an average write error rate of less than 1.5 × 10 −4 (more details in Supplementary Note S 3 ). We designate logic ‘0’ and ‘1’ to the parallel (P) low resistance state and anti-parallel (AP) high resistance state of MTJ, respectively.
Two-input logic operations are studied. The output cell is first initialized by writing ‘0’ to it. Then two input cells are connected to the output cell through the LL by turning on the corresponding transistors. Voltage pulses of amplitude of V logic , V logic , and 0, are simultaneously applied to the two input cells and the output cell, respectively. This is the same as grounding the output cell while applying a voltage pulse of V logic to the two input cells. Then, depending on the input cells’ states, the output cell will have a certain probability of being switched from ‘0’ to ‘1’. Such a cycle of operations is repeated n times, and the statistical mean of the output logic state, < D out >, is obtained. The entire process is repeated for different V logic values and input states. The basis for logic operations in the CRAM is the state-dependent resistance of the input cells. These shift and displace the output cell’s switching probability transfer curve. As a result, the output cell switches state based on specific input states, thereby implementing a logic function such as AND, OR, NAND, NOR, or MAJ. A specific initial state of the output cell and V logic value corresponds to one of these logic gates 66 . The time duration or pulse width of the voltage pulse applied during a logic operation is expected to contribute to most of the time required to complete a logic operation. In the following, we use the term logic speed to generally refer to the speed of a logic operation. Logic speed is approximately inversely proportional to the time duration of the voltage pulse used during a logic operation.
The experimental results are shown in Fig. 4 a, b . Generally, for a given input state, < D out > increases with increasing V logic . The < D out > response curves are input state-dependent. The four input states can be divided into three groups:
The ‘00’ input state yields the lowest resistance at the two input cells, so the output cell switches from ‘0’ to ‘1’ first (with the lowest V logic ).
The ‘11’ input state yields the highest resistance at the two input cells, so the output cell switches from ‘0’ to ‘1’ last (with the highest V logic ).
The ‘01’ and ‘10’ input states both yield resistance that falls in between that of ‘00’ and ‘11’so that the output cell’s response curve falls in between that of ‘00’ and ‘11’.
a Output logic average, D out , vs. logic voltage, V logic . In a 2-input logic operation, two input cells and one output MTJ cell are involved. The output cell’s terminal is grounded, while the common line is left floating. A logic operation voltage pulse is applied to the two input cells’ terminals for a fixed duration (pulse width) of 1 ms. Before each logic operation, input data is written to the input cells. After each logic operation, the output cell’s state is read. Each curve corresponds to a specific input state. Each data point represents the statistical average of the output cell’s logic state, < D out >, sampled by 1000 repeats ( n = 1000) of the operations. The separation between the < D out > curves indicates the margins for NOR or NAND operation, highlighted in blue and red, respectively. b Accuracy of 2-input NAND operation vs. logic voltage, V logic . The results in a can be converted into a more straightforward metric, accuracy, for the NAND truth table. The curve labeled ‘mean’ and ‘worst’ indicates the average and the worst-case accuracy across all input states, respectively. So, for NAND operation, the optimal logic voltage is indicated in such a plot where the mean or worst accuracy is maximized. c , d Accuracy of MAJ3 ( c ) and MAJ5 ( d ) logic operations vs. logic voltage, V logic . Each curve corresponds to an input state or a group of input states. And each data point represents the statistical average of the output MTJ logic state sampled by n = 1000 and n = 250, for c and d , respectively.
Figure 4a shows the experiment results. The two regions highlighted in blue and red that fall in between the three groups of response curves are suitable for NOR and NAND operations, respectively. For example, in the red region, the ‘11’ input has a high probability of yielding a ‘0’ output, while the other three input states have a high probability of yielding a ‘1’ output. This matches the expected truth table for a NAND logic gate. Therefore, if V logic is chosen carefully—within the red region for the CRAM 2-input logic operation—the operation performed is highly likely to be NAND.
The experimental results of < D out > can be converted into a straightforward format representing the accuracy for a specified logic function. This translation can be computed by simply subtracting < D out > from 1 for those input states where a ‘0’ output is expected in the truth table of the logic function. Figure 4b shows the NAND accuracy of the same 2-input CRAM logic operation. The ‘mean’ and ‘worst’ plots are based on the average value and minimum value of the accuracy, respectively, across all input state combinations at a fixed value for V logic . Based on the experimental results, if V logic = 0.624 or 0.616 V, the CRAM delivers a NAND operation with a best mean and a worst-case accuracy of about 99.4% and 99.0%, respectively. From a circuit perspective, both increasing the effective TMR ratio of input cells and/or making the output cell’s response curve steeper would increase the vertical separation of these input state-dependent curves, resulting in higher accuracy. For example, a higher effective TMR ratio of input cells results in a larger contrast of current in the output cell between different input states. Therefore, there is more ‘horizontal’ room to separate the < D out > curves associated with different input states so that for the inputs with which the output is expected to be ‘0’ or ‘1’, the < D out > of the output cell is closer to the expected value (‘0’ or ‘1’). Also note that for a logic operation, the ‘accuracy’ and ‘error rate’ are essentially two quantities describing the same thing: how true the logic operation is, statistically. By definition, the sum of accuracy and error rate is always 1. The higher or closer to 1 the accuracy is, the better. The lower or closer to 0 the error rate is, the better. Lastly, to facilitate better visualization of how the resistance changes of different input cell states are translated into voltage differences on the output cell resulting in it being switched or unswitched, we list the equivalent resistance of the two input cells combined in parallel and the resulting voltage on the output cell as follows: With V logic = 0.620 V, the equivalent resistance of input cells and the resulting voltage on the output cell are 0.4133 V and 1120 Ω, 0.3753 V, and 1461 Ω, and 0.3248 V and 2037 Ω, for input states ‘00’, ‘01’ or ‘10’, and ‘11’, respectively. Note that these values are estimated by the experiment-based modeling, which is introduced in the later part of this paper.
With more input cells, we also studied 3-input and 5-input majority logic operations. Figure 4c shows the accuracy of a 3-input MAJ3 logic operation. At V logic = −0.464 V, both the optimal mean and the worst-case accuracy are observed to be 86.5% and 78.0%, respectively. Similarly, for a 5-input MAJ5 logic operation, shown in Fig. 4d , both the optimal mean and the worst-case accuracy are observed to be 75% and 56%, respectively. As expected, comparing 2-input, 3-input, and 5-input logic operations, the accuracy decreases with an increasing number of inputs (more discussions and explanations in Supplementary Note S 4 ).
Having demonstrated fundamental elements of CRAM—memory write operations, memory read operations, and logic operations—we turn to more complex operations. We demonstrate a 1-bit full adder. This device takes two 1-bit operands, A and B, as well as a 1-bit carry-in, C, as inputs, and outputs a 1-bit sum, S, and a 1-bit carry-out, C out . A variety of implementations exist. We investigate two common designs: (1) one that uses a combination of majority and inversion logic gates, which we will refer to as a ‘MAJ + NOT’ design; and (2) one that uses only NAND gates, which we will refer to as an ‘all-NAND’ design. Figures 5 a and 5b illustrate these designs. Supplementary Note S 5 provides more details.
a , b Illustrations of the ‘MAJ + NOT’ and ‘all-NAND’ 1-bit full adder designs. Green and orange letter symbols indicate input and output data for the full adder, respectively. From left to right, numbered by ‘logic step,’ each drawing shows the intended input (green rectangle) and output (orange rectangle) cells involved in the logic operation. The text in purple under each drawing indicates the intended function of the logic operation (MAJ3, NAND, or MAJ5). c – f Experimental ( c , d ) and simulation ( e , f ) results of the output accuracy of 1-bit full adder operations by CRAM with the MAJ + NOT ( c , e ) and all-NAND ( d , f ) designs. The CRAM adder’s outputs, S and C out , are assessed against the expected values, i.e., their truth table, for all input states of A, B, and C. The accuracy of each result for each input state is shown by the numerical value in black font, as well as, represented by the color of the box with red (or blue) indicating wrong (or correct), or accuracy of 0% (100%). The accuracy is calculated based on the statistical average of outputs obtained by repeating the full adder execution n times, for n = 10,000. The experimental results for the MAJ + NOT ( c ) and all-NAND ( d ) designs are obtained by repeatedly executing the operation for all input states and observing the output states. The simulation results for the MAJ + NOT ( e ) and all-NAND ( f ) designs are obtained with probabilistic modeling, using Monte Carlo methods. The accuracy of individual logic operations is set to what was observed experimentally.
Figure 5c, f shows the experimental and simulation results for the MAJ + NOT and the all-NAND designs, respectively. Each plot is a colormap that lists the accuracy of the output bits S and C out , with each input state coded as [ABC]. The blue (red) indicates good/desired (bad/undesired) accuracy. In the boxes of colormap, results in saturated blue are the most desirable. The numerical values of accuracy are also labeled accordingly. From the experimental results for the MAJ + NOT design full adder shown in Fig. 5c , we make two observations:
The accuracy of C out is generally higher than that of S. This is because C out is directly produced by the first MAJ3 operation from inputs A, B, and C, while S is produced after additional logic operations. We also note that since C out is produced earlier than S, it is less impacted by error propagation and accumulation during each step; and the MAJ5 involved in producing S is inherently less accurate than the MAJ3.
Both C out and S have higher accuracy when the input [ABC] = 000 or 111 than in the other cases. This is expected since the input states of all ‘0’s and all ‘1’s yield higher accuracy than those with mixed numbers of ‘0’s and ‘1’s for both MAJ3 and MAJ5.
The experimental results for the all-NAND design are shown in Fig. 5d . The same observations regarding accuracy vs. inputs as the MAJ + NOT design apply. However, it is clear that the accuracy of the all-NAND full adder, at 78.5%, is higher than that of the MAJ + NOT full adder, at 63.8%. This is likely due to the fact that 2-input NAND operations are inherently more accurate than MAJ3 and MAJ5 operations. This offsets the impact of the additional steps required in the all-NAND design. We note that the accuracy of all computation blocks will improve as the underlying MTJ technology evolves. Accordingly, the relative accuracy of the all-NAND versus the MAJ + NOT designs may change 66 .
To understand the origin of errors, how they accumulate, and how they propagate, we performed numerical simulations of the full adder designs. These are based on probabilistic models of logic operations, implemented using Monte Carlo methods. Figure 5 e, f shows the simulation results for the MAJ + NOT and all-NAND designs, respectively. In these simulations, the accuracy of individual logic operations was set to match what was experimentally observed. The simulation results for the overall designs of the full adders correspond well to what was observed experimentally for these, confirming the validity of the proposed probabilistic models (more details in the Methods section and Supplementary Note S 6 ).
We note that beyond the inherent inaccuracy of logic operations, other factors such as device drift and device-to-device variation in MTJ devices will contribute to error in a CRAM. Specifically, drifts in temperature, external magnetic field, MTJ anisotropy, and MTJ resistance can lead to drift of the response curve, < D out >. Most likely, any such drift will result in a reduction (increase) of accuracy (error rate). More discussion regarding device-to-device variation is provided in Supplementary Note S 7 .
On the other hand, the accuracy of logic operations will significantly benefit from improvements in TMR ratio as MTJ technology evolves. To project the future accuracy of CRAM operations, we employ various types of physical modeling informed by existing experimental results (more details are provided in the Methods section and Supplementary Note S 8 ).
Three sets of assumptions on the accuracies (or error rates) of NAND logic operations underlie the following studies.
The ‘experimental’ assumptions are based on the best accuracy experimentally observed among the 9 NAND steps involved with the all-NAND 1-bit full adder. These are adjusted linearly to ensure that the error for inputs ‘01’ and ‘10’ equals that for input ‘11’. In reality, as supported by the experimental results shown in Fig. 4a , such a condition can be reached by properly tuning the V logic . Therefore, assuming the gate-level error rate is already optimized by tuning the V logic , then the per-input state NAND accuracies can be further simplified so that an error rate, δ (0 ≤ δ ≤ 1), can be used to characterize the error, accuracy, and probabilistic truth table of NAND operations in a CRAM. The NAND accuracy is [1, 1–δ, 1–δ, 1–δ], and the NAND probabilistic truth table is [1, 1– δ , 1– δ , δ ], both being a function of δ. Through the above-mentioned modeling and calculations, the ‘experimental’ assumptions yield δ = 0.0076, which corresponds to a TMR ratio of approximately 109%, based on experiments.
Two additional sets of assumptions, labeled as ‘production’ and ‘improved’, assume MTJ TMR ratios of 200% and 300%, respectively. These two assumptions yield δ = 2.1 × 10 −4 , and δ = 7.6 × 10 −6 , respectively, based on modeling and calculations. The ‘production’ assumptions represent the current industry-level TMR ratios developed for STT-MRAM technologies. The ‘improved’ assumptions present reasonable expectations for near-future MTJ developments.
CRAM NAND error rates vs. TMR ratio with various logic voltage pulse widths are shown in Fig. 6a . Higher TMR ratios and faster logic speed—so shorter V logic pulse widths—lead to smaller error rates. Further details can be found in Supplementary Note S 8 and in Supplementary Figure S 5 . Also included is an analysis of error rates vs. effective TMR ratio, which is independent of the specific TMR modeling. Note that, for all subsequent results, we will use the NAND error rate at the assumed TMR ratios, with pulse widths of 1 ms. This is more conservative but is consistent with the pulse widths used in the experimental results reported above.
a NAND gate minimum error rate vs. MTJ TMR ratio with various V logic pulse widths. For a given TMR ratio, the error rate is a function of V logic . So, the ‘minimum error rate’ represents the minimum error rate achievable with an appropriate V logic value. All subsequent studies use the error rates observed with 1 ms pulse widths (to be consistent with the earlier experimental studies) at assumed TMR ratios. b The NED error of a 4-bit dot-product matrix multiplier vs. TMR ratio. TMR ratios of 109%, 200%, and 300% are adopted for the ‘experimental,’ ‘production,’ and ‘improved’ assumptions, respectively. The size of the input matrix is indicated in the legend of the plot.
With these defined sets of assumptions, we provide projections of CRAM accuracy at a larger scale for meaningful applications. First, we evaluate ripple-carry adders and array multipliers 72 operating on scalar operands, with up to 6 bits. To evaluate the results, we adopt the normalized error distance (NED) metric 73 to represent the error of these primitives, as it has been shown to be more suitable for arithmetic primitives in the presence of computational error. We will refer to the error for a given primitive as ‘NED error’. We also define a complementary metric of ‘NED accuracy’ as the NED subtracted from 1 and then multiplied by 100%, to facilitate a more intuitive visualization of the error values. While the ‘experimental’ assumptions with a TMR ratio of 109% yield good overall accuracy for adders and multipliers, as the TMR ratio increases, the ‘production’ assumption with a TMR ratio of 200%, and the ‘improved’ assumption with a TMR ratio of 300%, yield significantly better or higher accuracies. Specifically, a 4-bit adder produces NED error of 2.8 × 10 −2 , 8.6 × 10 −4 , and 3.3 × 10 −5 , or NED accuracy of 97.2%, 99.914%, and 99.9967%, for the ‘experimental’, ‘production’, and ‘improved’ assumptions, respectively. A 4-bit multiplier produces NED error of 5.5 × 10 −2 , 1.8 × 10 −3 , and 6.6 × 10 −5 , or NED accuracy of 94.5%, 99.82%, and 99.9934%, for the three sets of assumptions, respectively. It is expected that when comparing the adder to the multiplier, since the latter is more complex and involves more gates, its accuracy is generally lower than that of the adder. Similarly, as the bit width of the adder or multiplier increases, their accuracy decreases. Further details and results with bit width up to 6-bit are provided in the Methods section and in Supplementary Note S 9 .
Then, using these primitives, we evaluate dot-product operations, which form the basis of matrix multiplication. They are heavily employed in many applications in both conventional domains and machine intelligence. Dot products consist of element-wise multiplication of two unsigned integer vectors, followed by addition. We perform additions with binary trees to maintain smaller circuit depth. Figure 6b shows the NED error of a 4-bit 4 × 4 dot-product matrix multiplier with respect to various TMR ratio assumptions. Like the adders and multipliers, as the TMR ratio increases, the NED error decreases, or the NED accuracy improves. Specifically, a 4-bit 4 × 4 dot-product matrix multiplier produces an NED error of 0.11, 3.4 × 10 −3 , and 1.2 × 10 −4 , or NED accuracy of 89%, 99.66%, and 99.988%, for the ‘experimental’, ‘production’, and ‘improved’ assumptions, respectively. Also, when comparing different input sizes (e.g., 1 × 1 to 4 × 4), as expected, the NED error is larger for larger input sizes due to the increased number of gates involved. Further details and results with bit width up to 5-bit are provided in the Methods section and in Supplementary Note S 9 .
To summarize the experimental work, an MTJ-based 1 × 7 CRAM array hardware was experimentally demonstrated and systematically evaluated. The basic memory write and read operations of CRAM were achieved with high reliability. The study on CRAM logic operations began with 2-input logic operations. It was found that a 2-input NAND operation could be performed with accuracy as high as 99.4%. As the number of input cells was increased, for example, for 3-input MAJ3 and 5-input MAJ5 operations, the accuracy decreased to 86.5% and 75%, respectively. The decrease was attributed to having too many levels corresponding to the input states crowding a limited operating margin. Next, two versions of a 1-bit full adder were experimentally demonstrated using the 1 × 7 CRAM array: an all-NAND version and a MAJ + NOT version. The all-NAND design achieved an accuracy of 78.5%, while the seemingly simpler MAJ + NOT, which involves 3- and 5-input MAJ operations, only achieved an accuracy of 63.8%. Note that although each type of logic operation achieves optimal accuracy performance with a specific voltage value, the value is expected to only need to be static or constant. Therefore, only a finite number of power rails is needed to accommodate the logic operations of the CRAM array. Also, if the multiple logic pulse duration is allowed by a peripheral design, it is possible to operate the CRAM array with a single set of power rails for both memory write and logic operations.
A probabilistic model was proposed that accounts for the origin of errors, their propagation, and their accumulation during a multi-step CRAM operation. The model was shown to be effective when matched with the experimental results for the 1-bit full adder. The working principles of this model were adopted for the rest of the studies.
A suite of MTJ device circuit models was fitted to the existing experimental data and used to project CRAM NAND gate-level accuracy in the form of error rates. The gate-level error rates were shown to be 7.6 × 10 −6 , with reasonable expectations of TMR ratio improvement as MTJ technology develops. Other device properties, such as the switching probability transfer curve, could also significantly affect the CRAM gate-level error rate. This calls for improvements or new discoveries of the physical mechanisms for memory read-out and memory write. Error is an inherent property of any physical hardware, including CMOS logic components, which are commonly perceived as deterministic. As the development of CRAM proceeds, the gate-level error rate of CRAM will be further reduced over time. For now, while the error rate of CRAM is still higher compared to that of CMOS logic circuits, CRAM is naturally more suitable for applications that require less precision but can still benefit from the true-in-memory computing features and advantages of CRAM, instead of those that require high precision and determinism. Additionally, logic operations with many inputs, such as majority, may be desirable in certain scenarios. And yet, these were shown to have lower accuracy than 2-input operations. So, a tradeoff might exist.
Lastly, building on the experimental demonstration and evaluation of the 1-bit full adder designs, simulation and analysis were performed for larger functional circuits: scalar addition and multiplication up to 6 bits and matrix multiplication up to 5 bits with input size up to 4 × 4. These are essential building blocks for many conventional and machine intelligence applications. The parameters for the simulations were experimentally measured values as well as reasonable projections for future MTJ technology. The results show promising accuracy performance of CRAM at a functional building block level. Furthermore, as device technologies progress, improved performance or new switching mechanisms could further reduce the gate-level error rate of CRAM. Error correction techniques could also be employed to suppress CRAM gate errors.
In summary, this work serves as the first step in experimentally demonstrating the viability, feasibility, and realistic properties of MTJ-based CRAM hardware. Through modeling and simulation, it also lays out the foundation for a coherent view of CRAM, from the device physics level up to the application level. Prior work had established the potential of CRAM through numerical simulation only. Accordingly, there had been considerable interest in the unique features, speed, power, and energy benefits of the technology. This study puts the earlier work on a firm experimental footing, providing application-critical metrics of gate-level accuracy or error rate and linking it to the application accuracy. It paves the way for future work on large-scale applications, in conventional domains as well as new ones emerging in machine intelligence. It also indicates the possibility of making competitive large-scale CMOS-integrated CRAM hardware.
The MTJ thin film stacks were grown by magnetron sputtering in a 12-source deposition system with a base pressure of 5 × 10 −9 Torr. The MgO barrier was fabricated by RF sputtering, while all the metallic layers were fabricated by DC sputtering. The stack structure is Si/SiO 2 /Ta(3)/Ru(6)/Ta(4)/Mo(1.2)/Co 20 Fe 60 B 20 (1)/MgO(0.9)/Co 20 Fe 60 B 20 (1.4)/Mo(1.9)/Ta(5)/Ru(7), where numbers in brackets indicate the thickness of the layer in nm. The stack was then annealed at 300 °C for 20 minutes in a rapid thermal annealing system under an Ar atmosphere (more information on the MTJ stack fabrication can be found in refs. 74 , 75 ).
The MTJ stacks were fabricated using three rounds of lithography similar to those described in ref. 76 . First, the bottom contacts were defined using photolithography followed by Ar+ ion milling etching. Then, the MTJ pillars were patterned into 120-nm circular nano-pillars via E-beam lithography and etched through Ar+ ion milling. After etching, SiO 2 was deposited via plasma-enhanced chemical vapor deposition (PECVD) to protect the nano-pillars. Finally, the top contacts were defined using photolithography, and the metallic electrodes of Ti (10 nm)/Au (100 nm) were deposited using electron beam evaporation.
The die of the MTJ array was diced into smaller pieces, with each piece containing about 10 MTJ devices. Each of the small pieces was mounted on a cartridge board, and up to 8 MTJ devices were wire-bonded to the electrodes of the cartridge board. Seven cartridge boards were inserted into the connection board, providing MTJs to the CRAM. The MTJ in each CRAM cell is selected among up to 8 MTJs on the corresponding cartridge board. In total, seven MTJs are selected from up to 56 MTJs. This method allows the user to find a collection of seven MTJs with minimum device-device variations.
An individual bias magnetic field was implemented for each of the seven MTJs on the connection board by positioning a permanent magnet at a certain distance from the MTJ devices. The bias magnetic field was used to compensate for intrinsic magnetic exchange bias and stray fields in the MTJ devices, thereby restoring the balance between the P and AP states. Additionally, slight rotation of bias field in the device plane was used to effectively adjust the switching voltage of each MTJ. More details can be found in Supplementary Note S 2 .
The connection board with seven MTJs was connected to the main board. On the main board, necessary active and passive electronic components were populated on the custom-designed PCB. The CRAM demo hardware circuit implemented a 1 × 7 CRAM array with a modified architecture to emphasize logic operations while compromising on memory operations bandwidth for simplicity. It was modified from the full-fledged 2T1M 40 architecture. It was equivalent to a 2T1M CRAM in logic mode, but it only had serial access to all cells for memory read and write operations (more details in Supplementary Note S 1 ). The hardware was powered by a battery and communicated with the controller PC wirelessly via Bluetooth®. In this way, the entire hardware was electrically isolated from the environment so that the risk of ESD to these sensitive MTJs was minimized.
The experiment control software running on a PC was implemented using National Instruments’ LabVIEW™. It was responsible for real-time measurements and control of the experiments, as well as necessary visualizations. Certain results were further analyzed post-experiment.
The simulation studies of accuracy as well as error origination, accumulation, and propagation began with a simple probabilistic model of each NAND logic operation. A probabilistic truth table was used to describe the expected statistical average of the output logical state. Then, the 1-bit full adder designs and operations were simulated using the Monte Carlo method with assumed probabilistic truth tables for each of the logic steps (see Supplementary Note S 6 ).
The experiment-based physics modeling and calculations for obtaining the projected CRAM logic operation accuracies began with an MTJ resistance-voltage model 77 , which was fit to the experimental data of TMR vs. bias voltage. The coefficients of this model were scaled accordingly to model projected TMR ratios higher than those observed experimentally. Then, a thermal activation model 78 , 79 of MTJ switching probability was fit to experimental data and was used to calculate the switching probability of the output MTJ cell under various bias voltages. Finally, the average of the output state, < D out >, could be calculated under various V logic , and the optimal NAND accuracies could be obtained in a manner similar to that discussed with Fig. 4 (more details in Supplementary Note S 8 ).
Further simulation studies of a ripple-carry adder, a systolic multiplier, and the dot-product operation of a matrix multiplication for various numbers of bits as well as matrix sizes were carried out using the same methods. More details can be found in Supplementary Note S 9 .
The authors declare that the data supporting the findings of this study are available within the paper and its supplementary information files.
Lecun, Y., Bengio, Y. & Hinton, G. Deep learning. Nature 521 , 436–444 (2015).
Article ADS CAS PubMed Google Scholar
Jordan, M. I. & Mitchell, T. M. Machine learning: trends, perspectives, and prospects. Science 349 , 255–260 (2015).
Article ADS MathSciNet CAS PubMed Google Scholar
Adomavicius, G. & Tuzhilin, A. Toward the next generation of recommender systems: a survey of the state-of-the-art and possible extensions. IEEE Trans. Knowl. Data Eng. 17 , 734–749 (2005).
Article Google Scholar
Hinton, G. et al. Deep neural networks for acoustic modeling in speech recognition: the shared views of four research groups. IEEE Signal Process. Mag. 29 , 82–97 (2012).
Article ADS Google Scholar
Collobert, R. et al. Natural language processing (almost) from scratch. J. Mach. Learn. Res. 12 , 2493–2537 (2011).
Google Scholar
Krizhevsky, A., Sutskever, I. & Hinton, G. E. ImageNet classification with deep convolutional neural networks. Commun. ACM 60 , 84–90 (2017).
Oh, K. S. & Jung, K. GPU implementation of neural networks. Pattern Recognit. 37 , 1311–1314 (2004).
Strigl, D., Kofler, K. & Podlipnig, S. Performance and scalability of GPU-based convolutional neural networks. In 2010 18th Euromicro Conference on Parallel, Distributed and Network-based Processing 317–324 (IEEE, 2010).
Nurvitadhi, E. et al. Accelerating binarized neural networks: comparison of FPGA, CPU, GPU, and ASIC. In 2016 International Conference on Field-Programmable Technology (FPT) 77–84 (IEEE, 2017).
Sawada, J. et al. TrueNorth ecosystem for brain-inspired computing: scalable systems, software, and applications. In SC ’16: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis 130–141 (IEEE, 2016).
Jouppi, N. P. et al. In-datacenter performance analysis of a tensor processing unit. In Proceedings of the 44th Annual International Symposium on Computer Architecture 1–12 (ACM, 2017).
Chen, Y. H., Krishna, T., Emer, J. S. & Sze, V. Eyeriss: an energy-efficient reconfigurable accelerator for deep convolutional neural networks. IEEE J. Solid-State Circuits 52 , 127–138 (2017).
Yin, S. et al. A high energy efficient reconfigurable hybrid neural network processor for deep learning applications. IEEE J. Solid-State Circuits 53 , 968–982 (2018).
Borghetti, J. et al. Memristive switches enable stateful logic operations via material implication. Nature 464 , 873–876 (2010).
Chi, P. et al. PRIME: a novel processing-in-memory architecture for neural network computation in ReRAM-based main memory. In 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA) 27–39 (ACM, 2016).
Shafiee, A. et al. ISAAC: a convolutional neural network accelerator with in-situ analog arithmetic in crossbars. In 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA) , 14–26 (2016).
Hu, M. et al. Dot-product engine for neuromorphic computing: programming 1T1M crossbar to accelerate matrix-vector multiplication. In 2016 53nd ACM/EDAC/IEEE Design Automation Conference (DAC) 1–6 (IEEE, 2016).
Seshadri, V. et al. Ambit: in-memory accelerator for bulk bitwise operations using commodity DRAM technology. In 2017 50th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO) 273–287 (IEEE, 2017).
Yao, P. et al. Fully hardware-implemented memristor convolutional neural network. Nature 577 , 641–646 (2020).
Jung, S. et al. A crossbar array of magnetoresistive memory devices for in-memory computing. Nature 601 , 211–216 (2022).
Keckler, S. W., Dally, W. J., Khailany, B., Garland, M. & Glasco, D. GPUs and the future of parallel computing. IEEE Micro 31 , 7–17 (2011).
Bergman, K. et al. ExaScale Computing Study: Technology Challenges in Achieving Exascale Systems . www.cse.nd.edu/Reports/2008/TR-2008-13.pdf (2008).
Horowitz, M. Computing’s energy problem (and what we can do about it). In 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC) 10–14 (IEEE, 2014).
Kim, D., Kung, J., Chai, S., Yalamanchili, S. & Mukhopadhyay, S. Neurocube: a programmable digital neuromorphic architecture with high-density 3D memory. In 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA) , 380–392 (2016).
Huang, J. et al. Active-routing: compute on the way for near-data processing. In 2019 IEEE International Symposium on High Performance Computer Architecture (HPCA) 674–686 (IEEE, 2019).
Nair, R. et al. Active memory cube: a processing-in-memory architecture for exascale systems. IBM J. Res. Dev. 59 , 17:1–17:14 (2015).
Pawlowski, J. T. Hybrid memory cube (HMC). In 2011 IEEE Hot Chips 23 Symposium (HCS) 1–24 (IEEE, 2011).
Gao, M., Ayers, G. & Kozyrakis, C. Practical near-data processing for in-memory analytics frameworks. In 2015 International Conference on Parallel Architecture and Compilation (PACT) 113–124 (IEEE, 2015).
Gao, M., Pu, J., Yang, X., Horowitz, M. & Kozyrakis, C. TETRIS: scalable and efficient neural network acceleration with 3D memory. SIGARCH Comput. Arch. News 45 , 751–764 (2017).
Aga, S. et al. Compute caches. In 2017 IEEE International Symposium on High Performance Computer Architecture (HPCA) 481–492 (IEEE, 2017).
Zidan, M. A., Strachan, J. P. & Lu, W. D. The future of electronics based on memristive systems. Nat. Electron. 1 , 22–29 (2018).
Jeon, K., Ryu, J. J., Jeong, D. S. & Kim, G. H. Dot-product operation in crossbar array using a self-rectifying resistive device. Adv. Mater. Interfaces 9 , 2200392 (2022).
Article CAS Google Scholar
Matsunaga, S. et al. Fabrication of a nonvolatile full adder based on logic-in-memory architecture using magnetic tunnel junctions. Appl. Phys. Express 1 , 091301 (2008).
Hanyu, T. et al. Standby-power-free integrated circuits using MTJ-based VLSI computing. Proc. IEEE 104 , 1844–1863 (2016).
Li, S. et al. Pinatubo: a processing-in-memory architecture for bulk bitwise operations in emerging non-volatile memories. In 2016 53rd ACM/EDAC/IEEE Design Automation Conference (DAC) 1–6 (IEEE, 2016).
Kvatinsky, S. et al. Memristor-based material implication (IMPLY) logic: design principles and methodologies. IEEE Trans. Very Large Scale Integr. Syst. 22 , 2054–2066 (2014).
Kvatinsky, S. et al. MAGIC—memristor-aided logic. IEEE Trans. Circuits Syst. II Express Briefs 61 , 895–899 (2014).
Wang, J.-P. & Harms, J. D. General structure for computational random access memory (CRAM). US patent 14/259,568 (2015).
Gupta, S., Imani, M. & Rosing, T. FELIX: fast and energy-efficient logic in memory. In 2018 IEEE/ACM International Conference on Computer-Aided Design (ICCAD) 1–7 (IEEE, 2018).
Chowdhury, Z. et al. Efficient in-memory processing using spintronics. IEEE Comput. Archit. Lett. 17 , 42–46 (2018).
Gao, F., Tziantzioulis, G. & Wentzlaff, D. ComputeDRAM: in-memory compute using off-the-shelf DRAMs. In 2019 52nd Annual IEEE/ACM International Symposium on Microarchitecture (MICRO) 100–113 (IEEE, 2019).
Truong, M. S. Q. et al. RACER: Bit-pipelined processing using resistive memory. In MICRO-54: 54th Annual IEEE/ACM International Symposium on Microarchitecture 100–116 (ACM, 2021).
Žutić, I., Fabian, J. & Das Sarma, S. Spintronics: fundamentals and applications. Rev. Mod. Phys. 76 , 323–410 (2004).
Nikonov, D. E. & Young, I. A. Benchmarking of beyond-CMOS exploratory devices for logic integrated circuits. IEEE J. Explor. Solid-State Comput. Devices Circuits 1 , 3–11 (2015).
Lee, T. Y. et al. World-most energy-efficient MRAM technology for non-volatile RAM applications. In 2022 International Electron Devices Meeting (IEDM) 10.7.1–10.7.4 (IEEE, 2022).
Jan, G. et al. Demonstration of ultra-low voltage and ultra low power STT-MRAM designed for compatibility with 0x node embedded LLC applications. In 2018 IEEE Symposium on VLSI Technology 65–66 (IEEE, 2018).
Zhao, H. et al. Sub-200 ps spin transfer torque switching in in-plane magnetic tunnel junctions with interface perpendicular anisotropy. J. Phys. D. Appl. Phys. 45 , 025001 (2012).
Julliere, M. Tunneling between ferromagnetic films. Phys. Lett. A 54 , 225–226 (1975).
Parkin, S. S. P. et al. Giant tunnelling magnetoresistance at room temperature with MgO (100) tunnel barriers. Nat. Mater. 3 , 862–867 (2004).
Yuasa, S., Nagahama, T., Fukushima, A., Suzuki, Y. & Ando, K. Giant room-temperature magnetoresistance in single-crystal Fe/MgO/Fe magnetic tunnel junctions. Nat. Mater. 3 , 868–871 (2004).
Berger, L. Emission of spin waves by a magnetic mulitlayer traversed by a current. Phys. Rev. B 54 , 9353–9358 (1996).
Article ADS CAS Google Scholar
Slonczewski, J. C. Current-driven excitation of magnetic multilayers. J. Magn. Magn. Mater. 159 , L1–L7 (1996).
Wei, L. et al. A 7Mb STT-MRAM in 22FFL FinFET technology with 4ns read sensing time at 0.9V using write-verify-write scheme and offset-cancellation sensing technique. In 2019 IEEE International Solid- State Circuits Conference - (ISSCC) 214–216 (IEEE, 2019).
Gallagher, W. J. et al. 22nm STT-MRAM for reflow and automotive uses with high yield, reliability, and magnetic immunity and with performance and shielding options. In 2019 International Electron Devices Meeting (IEDM) 2.7.1-2.7.4 (IEEE, 2019).
Chih, Y. Der et al. A 22nm 32Mb embedded STT-MRAM with 10ns read speed, 1M cycle write endurance, 10 years retention at 150 °C and high immunity to magnetic field interference. In 2020 IEEE International Solid- State Circuits Conference - (ISSCC) 222–224 (IEEE, 2020).
Edelstein, D. et al. A 14 nm embedded STT-MRAM CMOS technology. In 2020 International Electron Devices Meeting (IEDM) 11.5.1-11.5.4 (IEEE, 2020).
Chun, K. C. et al. A scaling roadmap and performance evaluation of in-plane and perpendicular MTJ based STT-MRAMs for high-density cache memory. IEEE J. Solid-State Circuits 48 , 598–610 (2013).
Lilja, D. J. et al. Systems and methods for direct communication between magnetic tunnel junctions. US patent 13/475,544 (2014).
Lyle, A. et al. Direct communication between magnetic tunnel junctions for nonvolatile logic fan-out architecture. Appl. Phys. Lett. 97 , 152504 (2010).
Zabihi, M. et al. Using spin-Hall MTJs to build an energy-efficient in-memory computation platform. In 20th International Symposium on Quality Electronic Design (ISQED) 52–57 (IEEE, 2019).
Currivan-Incorvia, J. A. et al. Logic circuit prototypes for three-terminal magnetic tunnel junctions with mobile domain walls. Nat. Commun. 7 , 1–7 (2016).
Alamdar, M. et al. Domain wall-magnetic tunnel junction spin-orbit torque devices and circuits for in-memory computing. Appl. Phys. Lett. 118 , 112401 (2021).
Zabihi, M. et al. Analyzing the effects of interconnect parasitics in the STT CRAM in-memory computational platform. IEEE J. Explor. Solid-State Comput. Devices Circuits 6 , 71–79 (2020).
Chowdhury, Z. I. et al. A DNA read alignment accelerator based on computational RAM. IEEE J. Explor. Solid-State Comput. Devices Circuits 6 , 80–88 (2020).
Chowdhury, Z. I. et al. CRAM-Seq: accelerating RNA-Seq abundance quantification using computational RAM. IEEE Trans. Emerg. Top. Comput. 10 , 2055–2071 (2022).
Zabihi, M. et al. In-memory processing on the spintronic CRAM: from hardware design to application mapping. IEEE Trans. Comput. 68 , 1159–1173 (2019).
Article MathSciNet Google Scholar
Cilasun, H. et al. CRAFFT: High resolution FFT accelerator in spintronic computational RAM. In 2020 57th ACM/IEEE Design Automation Conference (DAC) 1–6 (IEEE, 2020).
Resch, S. et al. PIMBALL: Binary neural networks in spintronic memory. ACM Trans. Archit. Code Optim. 16 , 41 (2019).
Chowdhury, Z. I. et al. CAMeleon: reconfigurable B(T)CAM in computational RAM. In Proceedings of the 2021 on Great Lakes Symposium on VLSI 57–63 (ACM, 2021).
Resch, S. et al. MOUSE: inference in non-volatile memory for energy harvesting applications. In 2020 53rd Annual IEEE/ACM International Symposium on Microarchitecture (MICRO) 400–414 (IEEE, 2020).
Lv, Y., Bloom, R. P. & Wang, J.-P. Experimental demonstration of probabilistic spin logic by magnetic tunnel junctions. IEEE Magn. Lett. 10 , 1–5 (2019).
Subathradevi, S. & Vennila, C. Systolic array multiplier for augmenting data center networks communication link. Cluster Comput. 22 , 13773–13783 (2019).
Liang, J., Han, J. & Lombardi, F. New metrics for the reliability of approximate and probabilistic adders. IEEE Trans. Comput. 62 , 1760–1771 (2013).
Almasi, H. et al. Perpendicular magnetic tunnel junction with W seed and capping layers. J. Appl. Phys. 121 , 153902 (2017).
Xu, M. et al. Voltage-controlled antiferromagnetism in magnetic tunnel junctions. Phys. Rev. Lett. 124 , 187701 (2020).
Lyu, D. et al. Sub-ns switching and cryogenic-temperature performance of mo-based perpendicular magnetic tunnel junctions. IEEE Electron Device Lett. 43 , 1215–1218 (2022).
Kim, J. et al. A technology-agnostic MTJ SPICE model with user-defined dimensions for STT-MRAM scalability studies. In 2015 IEEE Custom Integrated Circuits Conference (CICC) 1–4 (IEEE, 2015).
Diao, Z. et al. Spin-transfer torque switching in magnetic tunnel junctions and spin-transfer torque random access memory. J. Phys. Condens. Matter 19 , 165209 (2007).
Heindl, R., Rippard, W. H., Russek, S. E., Pufall, M. R. & Kos, A. B. Validity of the thermal activation model for spin-transfer torque switching in magnetic tunnel junctions. J. Appl. Phys. 109 , 073910 (2011).
Download references
This work was supported in part by the Defense Advanced Research Projects Agency (DARPA) via No. HR001117S0056-FP-042 “Advanced MTJs for computation in and near random-access memory” and by the National Institute of Standards and Technology. This work was supported in part by NSF SPX grant no. 1725420 and NSF ASCENT grant no. 2230963. The work at the University of Arizona is supported in part by NSF grant no. 2230124. The authors also thank Cisco Inc. for the support. Portions of this work were conducted in the Minnesota Nano Center, which was supported by the National Science Foundation through the National Nanotechnology Coordinated Infrastructure (NNCI) under Award No. ECCS-2025124. The authors acknowledge the Minnesota Supercomputing Institute (MSI, URL: http://www.msi.umn.edu ) at the University of Minnesota for providing resources that contributed to the research results reported within this paper. The authors thank Prof. Marc Riedel and Prof. John Sartori from the Department of Electrical and Computer Engineering at the University of Minnesota for proofreading the manuscript. Yang Lv, Brandon Zink, and Hüsrev Cılasun were CISCO Fellows.
Authors and affiliations.
Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN, 55455, USA
Yang Lv, Brandon R. Zink, Robert P. Bloom, Hüsrev Cılasun, Salonik Resch, Zamshed Chowdhury, Sachin S. Sapatnekar, Ulya Karpuzcu & Jian-Ping Wang
Department of Physics, University of Arizona, Tucson, Arizona, 85721, USA
Pravin Khanal, Ali Habiboglu & Weigang Wang
You can also search for this author in PubMed Google Scholar
J.-P.W. conceived the CRAM research and coordinated the entire project. Y.L. and J.-P.W. designed the experiments. Y.L. and R.P.B. designed and developed the demonstration hardware and software. P.K., A.H., and W.W. grew part of the perpendicular MTJ stacks. B.R.Z. fabricated the MTJ nanodevices. Y.L. conducted the CRAM demonstration experiments and analyzed the results. Y.L. studied the probabilistic model of CRAM operations and conducted simulations of a 1-bit full adder. Y.L., B.R.Z., and R.P.B. developed the device physics modeling of CRAM logic operations and gate-level error rates and conducted related calculations. H.C., S.R., Z.C., and U.K. carried out the simulation studies of the multi-bit adder, multiplier, and matrix multiplication. S.S. participated in discussions of modeling and simulation. All authors reviewed and discussed the results. Y.L. and J.-P.W. wrote the draft manuscript. All authors contributed to the completion of the manuscript.
Correspondence to Jian-Ping Wang .
Competing interests.
The authors declare no competing interests.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information, rights and permissions.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .
Reprints and permissions
Cite this article.
Lv, Y., Zink, B.R., Bloom, R.P. et al. Experimental demonstration of magnetic tunnel junction-based computational random-access memory. npj Unconv. Comput. 1 , 3 (2024). https://doi.org/10.1038/s44335-024-00003-3
Download citation
Received : 29 January 2024
Accepted : 29 May 2024
Published : 25 July 2024
DOI : https://doi.org/10.1038/s44335-024-00003-3
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research, free to your inbox weekly.
Click through the PLOS taxonomy to find articles in your field.
For more information about PLOS Subject Areas, click here .
Loading metrics
Open Access
Peer-reviewed
Research Article
Roles Conceptualization, Formal analysis, Investigation, Methodology, Project administration, Writing – original draft, Writing – review & editing
Affiliation Faculty of Business, School of Psychology, Justice and Behavioural Science, Charles Sturt University, Wagga Wagga, New South Wales, Australia
Roles Conceptualization, Methodology, Project administration, Supervision, Writing – original draft, Writing – review & editing
* E-mail: [email protected]
Videos glamourising disordered eating practices and body image concerns readily circulate on TikTok. Minimal empirical research has investigated the impact of TikTok content on body image and eating behaviour. The present study aimed to fill this gap in current research by examining the influence of pro-anorexia TikTok content on young women’s body image and degree of internalisation of beauty standards, whilst also exploring the impact of daily time spent on TikTok and the development of disordered eating behaviours. An experimental and cross-sectional design was used to explore body image and internalisation of beauty standards in relation to pro-anorexia TikTok content. Time spent on TikTok was examined in relation to the risk of developing orthorexia nervosa. A sample of 273 female-identifying persons aged 18–28 years were exposed to either pro-anorexia or neutral TikTok content. Pre- and post-test measures of body image and internalisation of beauty standards were obtained. Participants were divided into four groups based on average daily time spent on TikTok. Women exposed to pro-anorexia content displayed the greatest decrease in body image satisfaction and an increase in internalisation of societal beauty standards. Women exposed to neutral content also reported a decrease in body image satisfaction. Participants categorised as high and extreme daily TikTok users reported greater average disordered eating behaviour on the EAT-26 than participants with low and moderate use, however this finding was not statistically significant in relation to orthorexic behaviours. This research has implications for the mental health of young female TikTok users, with exposure to pro-anorexia content having immediate consequences for internalisation and body image dissatisfaction, potentially increasing one’s risk of developing disordered eating beliefs and behaviours.
Citation: Blackburn MR, Hogg RC (2024) #ForYou? the impact of pro-ana TikTok content on body image dissatisfaction and internalisation of societal beauty standards. PLoS ONE 19(8): e0307597. https://doi.org/10.1371/journal.pone.0307597
Editor: Barbara Guidi, University of Pisa, ITALY
Received: November 2, 2023; Accepted: July 8, 2024; Published: August 7, 2024
Copyright: © 2024 Blackburn, Hogg. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The data for this study can be found on Figshare via the following link: https://doi.org/10.6084/m9.figshare.25756800.v1 .
Funding: We acknowledge the financial support provided by Charles Sturt University.
Competing interests: The authors have declared that no competing interests exist.
Social media is a self-presentation device, a mode of entertainment, and a means of connecting with others [ 1 ], allowing for performance and the performance of identity [ 2 ], with social rewards built into its systems. Five to six years of the average human lifespan are now spent on social media sites [ 3 ] and visual platforms such as Instagram and TikTok increasingly dominate the cultural landscape of social media. Such visually oriented platforms are associated with higher levels of dysfunction in body image [ 4 ], while the COVID-19 pandemic has seen a rise in disordered eating behaviour [ 5 ]. Despite this, the field lacks a clear theoretical framework for understanding how social media usage heightens body image issues [ 6 ] and little research has specifically examined the impacts of TikTok based content. In this research, we sought to explore the impact of pro-anorexia TikTok content on body image satisfaction and internalisation of beauty standards for young women. The forthcoming sections of this literature review will highlight the features of social media content that may be particularly pernicious for young female users and will explore disordered eating and orthorexia in a social media context, concluding with a theoretical analysis of the relationship between social media and body image and internalisation of beauty standards, respectively.
Social media offers instant, quantifiable feedback coupled with idealised online imagery that may intersect with the value adolescents attribute to peer relationships and the sociocultural gender socialisation processes germane to this period of development, creating the “perfect storm” for young social media users, especially females [ 6 ]. In a study of 85 young, largely female eating disorder patients, a rise in awareness of online sites emphasizing thinness as beauty was evident from 2017 to 2020, with 60% of participants indicating that they knew of pro-ana websites and 22% of participants admitting to visiting them [ 7 ]. Research suggests that social media may also trigger those with extant eating disorders while simultaneously influencing healthy individuals to engage in disordered eating behaviour [ 8 ].
“Pro” eating disorder communities, hereafter referred to as “pro-ana” (pro-anorexia) communities, are a particular concern in a social media context. These communities encourage disordered eating, normalise disordered behaviours, and provide a means of connection for individuals who endorse anti-recovery from eating disorders [ 8 ]. Weight-loss tips, excessive exercise routines, and images of emaciated figures are routinely shared in these online communities [ 9 ], with extant research highlighting the association between viewing eating disorder content online and offline eating disorder behaviour [ 8 ]. Women who view pro-ana websites display increased eating disturbances, lowered body satisfaction, an increased drive for thinness, and higher levels of perfectionism when compared to women who have not viewed pro-ana content [ 10 , 11 ]. In research on adolescent girls, Stice [ 12 ] investigated the influence of exposure to media portraying the “thin-ideal” and found that perceived pressure to be thin was a predictor of increased body image dissatisfaction, which in turn led to increases in disordered eating behaviour. In similar research, Green [ 10 ] found that individuals with diagnosed eating disorders reported worsening symptoms after just 10-minutes of exposure to pro-ana content on the online platform, Tumblr.
The most downloaded social application (app) of 2021, TikTok is a social media platform that allows short-form video creation and sharing within a social media context [ 13 ]. Since its launch in 2017, TikTok has had over two billion downloads and has an estimated one billion users, the vast majority of which are children and teenagers [ 14 ]. Unlike other social media platforms where users have greater autonomy over the content generated on their homepage newsfeed, TikTok’s algorithm records data from single users and proposes videos designed to catch a user’s attention specifically, by creating a personalised “For You” page [ 15 ]. This feed will suggest videos from any creator on the platform, not just followed accounts. As such, if a user ‘interacts’ with a video, such as liking, sharing, commenting, or searching for related content, the algorithm will continue to produce similar videos on their “For You” page. The speed with which TikTok content can be created and consumed online may also be key to its impact. Any given social media user could watch more than a thousand videos on TikTok in an hour, creating a reinforcing effect that may have more impact than longer form content from a single creator [ 2 ].
Whilst the popularity of TikTok’s “For You” page has prompted global leaders in social media to build their own recommended content features, this feature remains most pronounced on TikTok. The “For You” page is the homepage of TikTok where users spend the majority of their time, compared to other social media platforms where homepages consist of a curation of content from followed accounts. Instagram’s explore page continues to emphasise established influencer culture and promote accounts of public figures or influencers with large followings. Contrastingly, TikTok’s unique algorithm makes content discoverability an even playing field, as any user’s content has the potential to reach a vast audience regardless of follower count or celebrity status. TikTok users therefore have less control over their homepage newsfeed compared to other social media platforms where users elect who they follow.
Unlike other social media platforms that implicitly showcase body ideals, TikTok contains explicit eating disorder content [ 16 ], while the “For You” page means that simply interacting with health and fitness videos can lead to unintended exposure to disordered eating content. Even seemingly benign “fitspiration” content may have psychological consequences for viewers. Beyond explicit pro-ana content, #GymTok and #FoodTok are two popular areas of content that provide a forum for users to create and consume content around their and others’ daily eating routines, weight loss transformations, and workout routines [ 2 ]. TikTok also frequently features content promoting clean eating, detox cleanses, and limited ingredient diets reflective of the current “food as medicine” movement of western culture [ 17 ], otherwise known as orthorexia. Despite efforts to ban such pro-ana related content, some videos easily circumvent controls [ 18 ], in part because many TikTok creators are non-public figures who are not liable to the backlash or cancellation that a public figure might receive for circulating socially irresponsible content.
Psychological analyses of eating disorders have historically focused on restrictive eating and the binge-purge cycle, however, more recently “positive” interests in nutrition have been examined. Orthorexia nervosa is characterised by a restrictive diet, ritualized patterns of eating, and rigid avoidance of foods deemed unhealthy or impure that consumes an individual’s focus [ 19 ]. Despite frequent observation of this distinct behavioural pattern by clinicians, orthorexia has received limited empirical attention and is not formally recognised as a psychiatric disorder [ 19 ]. Orthorexia and anorexia nervosa share traits of perfectionism, high trait anxiety, a high need to exert control, plus the potential for significant weight loss [ 19 ]. Termed ‘the disorder that cannot be diagnosed’ due to limited consensus around its features and the line between healthy and pathological eating practices, orthorexia mirrors the narrative of neoliberal self-improvement culture, wherein the body is treated as a site of performance and transformation.
Orthorexic restrictions and obsessions are routinely interpreted as signs of morality, health consciousness, and wellness [ 20 , 21 ]. Social media wellness influencers have played a significant role in normalising “clean [disordered] eating”. As one example, Turner and Lefevre [ 22 ] conducted an online survey of social media users following health food accounts and found that higher Instagram use was associated with a greater tendency towards orthorexia, with the prevalence of orthorexia among the study population at 49%, substantially higher than the general population (<1%). Similar health and food-related content on TikTok may provoke orthorexic tendencies among TikTok users, however, limited research has investigated orthorexic eating behaviour in the context of TikTok. The current study aims to bridge this gap in the literature around TikTok use and orthorexic tendencies. Disordered eating behaviour in the present study was measured by two separate but related constructs. ‘Restrictive’ disordered eating relates to dieting, oral control, and bulimic symptoms, whilst ‘healthy’ disordered eating constitutes orthorexic-like preoccupation with health food.
An established risk factor in the development and maintenance of disordered eating behaviour is negative body image. Body image is a multidimensional construct that represents an individual’s perceptions and attitudes about their physical-self and encompasses an evaluative function through which individuals compare perceptions of their actual “self” to “ideal” images [ 23 ]. This comparison may produce feelings of dissatisfaction about one’s own body image if a significant discrepancy exists between the actual and ideal self-image [ 23 ]. Body image is not necessarily congruent with actual physique, with research demonstrating that women categorised as having a healthy body mass index (BMI) nonetheless report dissatisfaction with their weight and engage in restrictive dietary behaviours to reduce their weight [ 24 ]. In addition, body image dissatisfaction is considered normative in Western society, particularly among adolescent women [ 25 ]. This may be attributable to the constant flow of media that exposes women to unrealistic images of thinness idealized within society [ 26 ].
One theoretical framework for understanding social media’s relationship with body image is the Social Comparison Theory, proposed by Festinger [ 27 ] who suggests that people naturally evaluate themselves in comparison to others via upward or downward social comparisons. Research supports the notion that women who frequently engage in maladaptive upward appearance-related social comparisons are more likely to experience body image dissatisfaction and disordered eating [ 25 , 28 ], while visual exposure to thin bodies may detrimentally modulate one’s level of body image satisfaction [ 29 – 31 ]. In their study of undergraduate females, Engeln-Maddox [ 29 ] found that participants made upward social comparisons to images of thin models which were strongly associated with decreases in body image satisfaction and internalisation of thinness. Similarly, Tiggemann [ 32 ] found that adolescents who spent more time watching television featuring attractive actors and actresses reported an increased desire for thinness, theorised to be a result of increased social comparison to attractive media personalities.
The Transactional Model [ 33 ] extends Social Comparison Theory by emphasising the multifaceted and complex nature of social media influences on body image. This model acknowledges that individual differences may predispose a person to utilise social media for gratification, and highlights that as time spent on social media increases, so too does body image dissatisfaction [ 33 ]. In line with this, a recent review of literature by Frieiro Padín and colleagues [ 34 ] indicated that time spent on social media was strongly correlated with eating disorder psychopathologies, as well as heightened body image concerns, internalisation of the thin ideal, and lower levels of self-esteem. Time on social media also correlated with heightened body image concerns to a far greater extent than general internet usage [ 35 , 36 ].
Body image ideals are not static. The traditional ideal of rib-protruding bodies from the 90s, known colloquially as “heroin chic”, have recently shifted to a celebration of the “slim-thicc” figure, consisting of a cinched, flat waist with curvy hips, ample breasts, and large behinds [ 37 ]. The “slim-thicc” aesthetic allows women to be bigger than previous body ideals, yet this figure is arguably more unattainable than the thin-ideal as surgical intervention is commonly needed to achieve it, depending on genetics and body type. The idealisation of the “slim-thicc” figure is highlighted by the “Brazilian butt lift” (BBL), a potentially life-threatening procedure that is nonetheless the fastest growing category of plastic surgery, doubling in growth over the past five years, despite the life-threatening potential of the procedure [ 38 ]. Research suggests that the slim-thicc ideal is no less damaging nor threatening of body image than the thin-ideal. Indeed, in experimental research on body ideals, McComb and Mills [ 39 ] found that the greatest body dissatisfaction levels in female undergraduate students were observed among those exposed to imagery of the slim-thicc physique, relative to that exhibited by those exposed to the thin-ideal and fit-ideal physique, as well as the control condition.
Recent body ideals have also favoured muscular thin presentations, considered to represent health and fitness as evident in the “#fitspiration” Instagram hashtag that features over 65 million images [ 40 ]. Fitspiration has the potential to positively influence women’s health and wellbeing by promoting exercise engagement and healthy eating, yet various content analyses of fitspiration images highlight aspects of fitspiration that warrant concern [see 40 , 41 ]. Notably, fitspiration typically showcases only one body type and women whose bodies do not meet this standard may experience body dissatisfaction [ 40 ], while the gamification of exercise, such as receiving likes for every ten sit-ups, segues with the intensive self-control and competitiveness that often underpins eating disorders and eating disorder communities [ 1 ].
In recent experimental research, Pryde and Prichard [ 42 ] examined the effect of exposure to fitspiration TikTok content on the body dissatisfaction, appearance comparison, and mood of young Australian women. Viewing fitspiration TikTok videos led to increased negative mood and increased appearance comparison but did not impact body dissatisfaction. This finding contradicts previous research and may be due to fitspiration content showcasing body functionality rather than aesthetic, which may lead to positive outcomes for viewers. The fitspiration content used by Pryde and Prichard [ 42 ] did not contain the harmful themes regularly found in other forms of fitspiration content. Appearance comparison was significant in the relationship between TikTok content and body dissatisfaction and mood, suggesting that this may be a key mechanism through which fitspiration content leads to negative body image outcomes and supporting the notion that fitspiration promotes a focus on appearance rather than health.
Body image dissatisfaction among women is associated with co-morbid psychological disturbances and the development of disordered eating behaviours [ 43 , 44 ]. A large body of research indicates that higher levels of both general and appearance-related social comparison are associated with disordered eating in undergraduate populations [ 10 , 28 , 45 – 48 ]. As one example, Lindner et al. [ 46 ] investigated the impact of the female-to-male ratio of college campuses on female students’ engagement in social comparison and eating pathology. Their findings lend support to the Social Comparison Theory, indicating that the highest levels of eating pathology and social comparison were found among women attending colleges with predominantly female undergraduate populations. A strong relationship was also found between eating pathology and engagement in appearance-related social comparisons independent of actual weight. Lindner et al. [ 46 ] surmised that these results suggest social comparison and eating pathology behaviours are due to students’ perceptual distortions of their own bodies, potentially fostered by pressures exerted from peers to be thin.
Similarly, Corning et al. [ 45 ] investigated the social comparison behaviours of women with eating disorder symptoms and their asymptomatic peers. Results illustrated that a greater tendency to engage in everyday social comparison predicted the presence of eating disorder symptoms, while women with eating disorder symptoms made significantly more social comparisons of their own bodies. Such findings are supported by subsequent research, with Hamel et al. [ 28 ] finding that adolescents with a diagnosed eating disorder engaged in significantly more body-related social comparison than adolescents diagnosed with a depressive disorder or no diagnosis. Body-related social comparison was also significantly positively correlated with disordered eating behaviours. While extant research has focused upon social comparison as it has occurred through traditional media outlets, less research has investigated the facilitation of social comparison through social media platforms, particularly contemporary platforms such as TikTok.
The extent to which one’s body image is impacted by images and messages conveyed by the media is determined by the degree to which these images and messages are internalised. Some may argue that social media platforms are distinct from what occurs in “real” life, creating fewer opportunities for internalisation to occur. Yet as Pierce [ 2 ] argues, platforms such as TikTok create their own realities, allowing users to explore their identities, form relationships, engage with culture and world events, and even develop new patterns of speech and writing. TikTok trends commonly infiltrate society, underscoring the impact of social media on life beyond the online world and thus a sociocultural analysis of TikTok is warranted. Sociocultural theories suggest that society portrays thinness as the ideal body shape for women, resulting in an internationalisation of the “thin is good” assumption for women. This in turn results in lowered body image satisfaction and other negative outcomes [ 43 ]. The significance of social influences, including the role of family, peers, and the media, is emphasised by sociocultural theory, with individuals more likely to internalise the thin ideal when they encounter pressuring messages that they are not thin enough from social influences [ 48 ]. Within this theoretical approach, an individual’s degree of thin ideal internalisation is theorised to depend on their acceptance of socially defined ideals of attractiveness and is reflected in their engagement in behaviours that adhere to these socially defined ideals [ 49 ].
Building on this, the tripartite influence model suggests that disordered eating behaviours arise due to pressure from social agents, specifically media, family, and peers. This pressure centres on conforming to appearance ideals and may lead to engagement in social comparison and the internalisation of thin ideals [ 48 ]. This is relevant in a digital context given social media provides endless opportunities for individuals to practice social comparison and for many users, social comparison on TikTok is peer-based as well as media-based. According to the tripartite model, social comparisons have been consistently associated with a higher degree of thin ideal internalisation, self-objectification, drive for thinness, and weight dissatisfaction [ 50 ]. Furthermore, and in contrast to traditional media where social agents are mainly models, celebrities, and movie stars, social agents on social media can include peers, friends, family, and individuals who have a relationship with the individual. Social media content generated by “everyday” people, rather than super models or movie stars, may result in comparisons that are more horizontal in nature. This is particularly evident on TikTok where content creators are rarely famous before creating a TikTok account, often remain micro-influencers after achieving some notoriety, and are usually around the same age as those viewing their content.
Pressure to be thin from alike peers may have a particularly pronounced impact on one’s degree of internalisation of the thinness ideal. Indeed, Stice et al. [ 51 ] found that after listening to young thin women complain about “feeling fat”, their adolescent participant sample reported increased body image dissatisfaction, suggesting that pressure from peers perpetuates the thinness ideal, leading to internalisation of the ideal and subsequent body dissatisfaction. Similarly, it was found that adolescent females were more likely to engage in weight loss behaviour if a high portion of peers with a similar BMI were also engaging in these behaviours, illustrating that appearance pressure exerted by alike peers may result in thin-ideal internalisation and engagement in weight loss behaviours to control body weight and shape [ 52 ]. Such findings raise questions around whether those most similar to us have the greatest impact upon thin-ideal internalisation, body image dissatisfaction, and disordered eating behaviours.
In further support for the tripartite influence model, research by Thompson et al. [ 48 ] indicates that the ideals promoted through social media trends are internalized despite being unattainable, resulting in body image dissatisfaction and disordered eating behaviour. Similarly, Mingoia et al. [ 53 ] found a positive association between the use of social networking sites and thin ideal internalisation in women, indicating that greater use of social networking sites was linked to significantly higher internalisation of the thin ideal. Interestingly, the use of appearance-related features (e.g., posting or viewing photographs or videos) was more strongly related with internalisation than the broad use of social networking sites (e.g., writing status’, messaging features) [ 53 ]. Correlational and experimental research alike has demonstrated that thin ideal internalisation is related to body image dissatisfaction and leads to expressions of disordered eating such as restrictive dieting and binge-purge symptoms [ 31 , 48 , 54 , 55 ]. Subsequent expressions of disordered eating may be seen as an attempt to control weight and body shape to conform to societal beauty standards of thinness [ 51 ].
This sociocultural perspective is exemplified by Grabe et al’s. [ 54 ] meta-analysis of research on the associations between media exposure to women’s body dissatisfaction, internalisation of the thin ideal, and eating behaviours and beliefs, illustrating that exposure to media images propagating the thin ideal is related to and indeed, may lead to body image concerns and increased endorsement of disordered eating behaviours in women. Similarly, Groesz et al. [ 55 ] conducted a meta-analysis to examine experimental manipulations of the thin beauty ideal. They found that body image was significantly more negative after viewing thin media images than after viewing images of average size models, plus size models, or inanimate objects. This effect size was stronger for participants who were more vulnerable to activation of the thinness schema. Groesz et al. [ 55 ] conclude that their results align with the sociocultural theory perspective that media promulgates a thin ideal that in turn provokes body dissatisfaction.
Existing research has established the relationship between body image dissatisfaction and disordered eating behaviours and social media platforms such as Instagram and Twitter. The unique implications of the TikTok ‘For You Page’, as well as the dominance of peer-created and explicit disordered eating content on TikTok suggests that the influence of this platform warrants specific consideration. This study adds to extant literature by utilising an experimental design to examine the influence of exposure to pro-ana TikTok content on body image and internalisation of societal beauty standards. A cross-sectional design was used to investigate the effect of daily TikTok and the development of disordered eating behaviours. Although body image disturbance and eating disorders are not limited to women, varying sociocultural factors have been implicated in the development of disordered eating behaviour in men and women [ 45 ], while issues facing trans people warrant specific consideration beyond the scope of this study, therefore the present sample contains only female-identifying participants.
The current study aimed to investigate the impact of pro-ana TikTok content on young women’s body image satisfaction and internalisation of beauty standards, as well as exploring daily TikTok use and the development of disordered eating behaviour. First, in line with the cross-sectional component of the study, it was hypothesized that women who spend greater time on TikTok per day would report significantly more disordered eating behaviour than women who spend low amounts of time on TikTok per day. Second, it was hypothesized that women in the pro-ana TikTok group would report a significant decrease in body image satisfaction state following exposure to the pro-ana content compared to women in the control group. Third, it was hypothesized that women in the pro-ana Tik Tok group would report increased internalisation of societal beauty standards following exposure to pro-ana TikTok content compared to women in the control group.
Participants in the current study included 273 women aged between 18 to 28 years sourced from the general population of TikTok users. The predominant country of residence of the sample was Australia, with 15 participants indicating they currently reside outside of Australia. Of the remaining data relating to the two conditions of the study, 126 participants were randomly allocated into the experimental condition, and 147 participants were randomly allocated into the control condition. Snowball sampling was used to recruit participants through social media, online survey sharing platforms, and word-of-mouth, with first-year University students targeted for recruitment by offering class credit in return for participation. Participants could withdraw their consent at any time by exiting the study prior to completion of the survey.
The current study employed a questionnaire set that included a demographic questionnaire, and five scales measuring disordered eating behaviour, body satisfaction, and internalisation of societal beauty standards, as well as perfectionism, the latter of which was not examined in the present study.
The demographic questionnaire required participants to answer a series of questions relating to their gender, age, relationship status, ethnicity, country of residence, TikTok usage, and exercise routine. A screening question redirected non-female-identifying persons from the study. Responses to the TikTok usage items were examined cross-sectionally with responses on the EAT-26 and ORTO15 used to examine the influence of daily TikTok use and the presentation of disordered eating behaviours.
The Eating Attitudes Test (EAT-26, [ 56 ]) is a short form of the original 40-item EAT scale [ 57 ] which measures symptoms and concerns characteristic of eating disorders. The 26-item short-form version of the EAT was utilised in the present study due to its established reliability and validity, and strong correlation with the EAT-40 [ 56 ].
Responses to the 26-items are self-reported using a 6-point Likert scale ranging from Always (3) to Never (0) [ 56 ]. The EAT-26 consists of three subscales including dieting, bulimia and food preoccupation, and oral control. Five behavioural questions are included in Part C of the EAT-26 to determine the presence and frequency of extreme weight-control behaviours including binge eating, self-induced vomiting, laxative usage, and excessive exercise [ 56 ]. Higher scores indicate greater disordered eating behaviour, and those with a total score of 20 or greater are, in clinical contexts, typically highlighted as requiring further assessment and advice of a mental health professional [ 56 ].
Internal consistency of the EAT-26 was established in initial psychometric studies which reported a Cronbach’s alpha of.85 [ 58 ]. For the current study, the Cronbach’s alpha = .91. Previous research has also demonstrated that the EAT-26 has strong test-retest reliability (e.g., 0.84) [ 59 ], as well as acceptable criterion-related validity for differentiating between eating disorder populations and non-disordered populations [ 56 ]. In the current study, the EAT-26 was used to measure disordered eating behaviour, and the cut-off score of 20 and above was adopted to categorise increased disordered eating behaviour. Given how this construct is measured, from this point forward the present study will refer to EAT-26 responses as ‘restrictive’ type disordered eating.
The ORTO-15 is a 15-item screening measure that assesses orthorexia nervosa risk through questions regarding the perceived effects of eating healthy food (e.g. “Do you think that consuming healthy food may improve your appearance?”), eating habits (e.g. “At present, are you alone when having meals?”), and the extent to which concerns about food influence daily life (e.g. “Does the thought of food worry you for more than three hours a day?”) [ 19 ]. Responses are self-reported using a 4-point Likert scale ranging from always , often , sometimes , or never . Individual items are coded and summed to derive a total score. Donini et al. [ 60 ] established a cut off total score of 40; scores below 40 indicate orthorexia behaviours, whilst scores 40 or above reflect normal eating behaviour. This cut off score was determined by Donini et al. [ 60 ] as their results revealed the ORTO-15 demonstrated good predictive capability at the threshold of 40 compared to other potential threshold values.
Although the ORTO-15 is the most widely accepted screening tool to assess orthorexia risk, it is still only partially validated [ 61 ], and inconsistencies of the measures’ reliability and validity exist in current literature. For example, Roncero et al. [ 62 ] estimated that the reliability of the ORTO-15 using Cronbach’s alpha was between 0.20 and 0.23, however, after removing certain items, the reliability coefficients were between 0.74 and 0.83. Contrastingly, Costa and colleagues’ [ 63 ] review of current literature surrounding orthorexia suggested adequate internal consistency (Cronbach’s alpha = 0.83 to 0.91) with all 15-items.
In the present study, a reliability analysis revealed unacceptable reliability for the ORTO-15 (α = .24). Principal components factor analysis identified two factors within the ORTO-15, one relating to dieting and the other to preoccupation with health food. Separate reliability analyses were performed on the items that comprised these two factors and the diet-related items did not have acceptable reliability (α = -.40), whilst the health food-related items bordered on acceptable reliability at α = .63. Consequently, only the health food-related items were retained in the current study following consideration of Pallant’s [ 64 ] assertion that Cronbach alpha values are sensitive to the number of items on a scale and it is therefore common to obtain low values on scales with less than ten items. Pallant [ 64 ] notes that in cases such as this, it is appropriate to report the inter-item correlation of the items, while Briggs and Cheek [ 65 ] advise an optimal range for the inter-item correlation between.2 to.4, with the health food-related items in the current study obtaining an inter-item correlation of.25. Throughout this study, the construct measured by these ORTO-15 items will be referred to as ‘healthy’ type disordered eating to reflect this obsessive health food preoccupation and differentiate between the two disordered eating dependent variables measured in the current study.
The Body Image States Scale (BISS) by Cash and colleagues [ 66 ] is a six-item measure of momentary evaluative and affective experiences of one’s own physical appearance. The BISS evaluates the following aspects of current body experience: dissatisfaction-satisfaction with overall physical appearance; dissatisfaction-satisfaction with one’s body size and shape; dissatisfaction-satisfaction with one’s weight; feelings of physical attractiveness-unattractiveness; current feelings about one’s looks relative to how one usually feels; and evaluation of one’s appearance relative to how the average person looks [ 66 ]. Participants responded to these items using a 9-point Likert-type scale which is presented in a negative-to-positive direction for half of the items, and a positive-to-negative direction for the other half [ 66 ]. Respondents were instructed to select the statement that best captured how they felt “ right now at this very moment ”. A total BISS score was calculated by reverse-scoring the three positive-to-negative items, summing the six-items, and finding the mean, with higher total BISS scores indicating more favourable body image states.
During the development and implementation of the BISS, Cash and colleagues [ 66 ] report acceptable internal consistency and moderate stability over time, an anticipated outcome due to the nature of the BISS as a state assessment tool. The BISS was also appropriately correlated with a range of trait measures of body image, highlighting its convergent validity [ 66 ]. Cash and colleagues [ 66 ] also report that the BISS is sensitive to reactions in positive and negative situational contexts and has good construct validity. An acceptable Cronbach’s alpha coefficient of.88 was obtained in the current study.
The Sociocultural Attitudes Towards Appearance Questionnaire– 4 (SATAQ-4) [ 67 ] is a 22-item self-report questionnaire that assesses the influence of interpersonal and sociocultural appearance ideals on one’s body image, eating disturbance, and self-esteem. Ratings are captured on a 5-point Likert scale which asks participants to specify their level of agreement with each statement by choosing from 1 ( definitely disagree ) through to 5 ( definitely agree ), with higher scores indicative of greater pressure to conform to, or greater internalisation of, interpersonal and sociocultural appearance ideals [ 67 ]. The five subscales of the SATAQ-4 measure: internalisation of thin/low body fat ideals, internalisation of muscular/athletic ideals, influence of pressures from family, influence of pressure from peers, and influence of pressures from the media [ 67 ]. For the purposes of the present study, the questions from the media pressure subscale were modified to enquire specifically about social media rather than traditional forms of media.
Across all samples in Schaefer et al’s. [ 67 ] study, the internal consistency of the five SATAQ-4 subscales is considered acceptable to excellent, with Cronbach’s alpha scores between 0.75 and 0.95. These subscales also displayed good convergent validity with other measures of body satisfaction, eating disorder risk, and self-esteem [ 67 ]. Pearson product-moment correlations between the SATAQ-4 subscales and convergent measures revealed medium to large positive associations with eating disorder symptomology, medium negative associations with body satisfaction, and small negative associations with self-esteem [ 67 ]. A Cronbach’s alpha of.87 was obtained in the present study, demonstrating acceptable internal consistency.
Ethical approval for the present study was granted by the Charles Sturt University Human Research Ethics Committee (Approval number H21155) prior to data collection. Participants were directed to the study via an online link to QuestionPro where they were provided an explanation of the study, their rights, and contact details of relevant support services if they were to become distressed. Participants gave informed consent by clicking on a link that read, “I consent to participate” at the beginning of the survey and then again through the submission of their completed survey. Any incomplete responses were not included in the dataset. Data collection commenced on the 30 th of July 2021 and ceased on the 1st of October 2021. In line with the cross-sectional and descriptive aspects of the research, participants were asked demographic questions about their gender, age, relationship status, ethnicity, country of residence, TikTok usage, and exercise habits. Participants then completed the experimental set in the following order: BISS (pre-test), SATAQ-4 (pre-test), EAT-26, ORTO-15, Experimental intervention (control or experimental TikTok video condition), SATAQ-4 (post-test), BISS (post-test), and debrief. All questionnaires presented to each participant were identical. Measures were not randomised to ensure that body image and internalisation were assessed at both pre- and post-test to evaluate the experimental manipulation.
Participants were randomly allocated to one of two conditions: experimental (pro-ana TikTok video) or control (“normal” TikTok video). Participants allocated to the experimental condition watched a compilation of TikTok videos containing explicit disordered eating messages such as young women restricting their food, displaying gallows humour about their disordered eating behaviour, starving themselves, and providing weight loss tips such as eating ice cubes and chewing gum to curve hunger. Participants in the experimental condition were also exposed to more implicit body image ideals typical of fitspiration-style content. This included thin women displaying their abdomens, cinched waists, dancing in two-piece swimwear, along with workout and juice cleanse videos promising fast weight loss. Participants in the control condition viewed a compilation of TikTok videos containing scenes relating to nature, cooking and recipes, animals, and comedy. After viewing the 7- to 8-minute TikTok video, all participants completed measures of internalisation and body satisfaction again to assess the influence of either the pro-ana TikTok video or the normal TikTok video. The debrief statement made explicit to participants the rationale of the study and explained the non-normative content of the videos shown to the experimental group. A small financial incentive was offered via a prize draw of five vouchers.
The data from QuestionPro was collated and analysed using IBM SPSS Statistics software, Version 28. All measures and manipulations in the study have been disclosed, alongside the method of determining the final sample size. No data collection was conducted following analysis of the data. Data for this study is available via the Figshare data repository and can be accessed at https://doi.org/10.6084/m9.figshare.25756800.v1 . This study was not preregistered. Sample size was determined before any data analysis. A priori power analyses were conducted using G*Power to determine the minimum sample sizes required to test the study hypotheses. Results indicated the required sample sizes to achieve 90% power for detecting medium effects, with a significance criterion of α = 0.05, were: N = 108 for the mixed between-within subjects ANOVAs and N = 232 for the one-way between groups ANOVAs. According to these recommendations, adequate statistical power was achieved. All univariate and multivariate assumptions were checked and found to be met. All scales and independent variables were normally distributed.
The analysis of the current study including data screening processes, descriptive statistics, and hypothesis testing will be presented in this section. Hypothesis testing began with two separate mixed between-within subjects analysis of variance models (ANOVAs) to examine the impact of the experimental manipulation on the independent variables of body image and internalisation of appearance ideals and pressure. Finally, the effect of time spent using TikTok daily on restrictive and ‘healthy’ disordered eating behaviour was explored cross-sectionally using two separate one-way between-subjects ANOVAs.
Prior to statistical analysis, data were screened for entry errors and missing data. Of the 838 participants who initially consented to participate in the survey, 555 responses were insufficiently complete for data analysis. As participants were permitted to withdraw their consent by exiting the online survey, these results were excluded from all subsequent analyses. Of those that did not complete the study, the majority withdrew during the BISS (pre-test) and the ORTO-15, suggesting that these participants potentially experienced discomfort or distress when asked to reflect on their appearance and their eating behaviours. Of the completed responses, nine were excluded due to not meeting the study’s stated age eligibility and another case was excluded due to disclosure of a previous eating disorder diagnosis. The remaining data set comprised of 273 participants.
Demographic characteristics..
In the current sample, 50% of participants reported being currently single and most participants (83%) were Caucasian, with 71% of participants indicating that they spent up to two hours per day using TikTok. Further demographic information is provided in Table 1 .
https://doi.org/10.1371/journal.pone.0307597.t001
Participants in the current study reported that entertainment (75%), fashion (59%), beauty/skincare (54%), cooking/recipes (51%) and life hacks/advice (51%) content frequently occurred on their For You page. Largely in keeping with this, participants reported experiencing the most enjoyment from viewing entertainment (84%), life hacks/advice (57%), home renovation (56%), recipes/cooking (56%), and fashion (54%) content on their For You page.
In the current sample, 64% of participants reported being exposed to disordered eating content via their For You page. Only 15% of participants had not been exposed to any negative content themes. Further descriptive For You page content information is displayed below in Table 2 .
https://doi.org/10.1371/journal.pone.0307597.t002
Notably, 43% of the participant sample were frequently exposed to fitness and sports related content and the same percentage of the sample enjoyed seeing this content, suggesting that content broadly aligned with #fitspiration was consumed and appreciated by nearly half of participants. Concerningly, 40–60% of participants had been exposed to negative TikTok content via the For You Page, with content ranging from self-harm and suicidality to violence and illegal activity. No data was collected on the specifics of this content, however, and it is possible that some “negative” content may be framed from a proactive, preventative perspective, and this warrants further consideration.
Hypothesis 1: daily tiktok use and disordered eating behaviour..
To test the cross-sectional analysis of this study, two separate one-way between-groups ANOVAs were conducted to explore the impact of daily amount of TikTok use on ‘healthy’ disordered eating and restrictive disordered eating behaviour. This was necessary as time on TikTok was measured categorically. Participants were divided into four groups according to their average daily time spent using TikTok (Low use group: 1 hour or less; Moderate use group: 1–2 hours; High use group: 2–3 hours; Extreme use group: 3+ hours). Homogeneity of variance could be assumed for each ANOVA as indicated by non-significant Levene’s Test Statistics.
There was no statistically significant difference at the p < .05 level in ORTO15 scores for the four TikTok usage groups: F (3, 269) = .38, p = .78, indicating that ‘healthy’ disordered eating did not significantly differ across women who use TikTok for different periods of time per day. The effect size, calculated using eta squared, was.004, which is considered small in Cohen’s [ 68 ] terms. This small effect size is congruent with the non-significant finding.
The second ANOVA measuring differences among EAT-26 scores across the four TikTok usage groups also yielded a non-significant result: F (3, 269) = 1.21, p = .31. Eta squared was calculated as.01, representing a small effect size [ 68 ] consistent with this non-significant result. The means and standard deviations of the four TikTok usage groups across dependent variables of ‘healthy’ and restrictive disordered eating, as measured by the ORTO15 and the EAT-26 respectively, are displayed in Table 3 .
https://doi.org/10.1371/journal.pone.0307597.t003
Hypothesis 2: body image satisfaction across groups from pre-test to post-test..
To evaluate the effect of the experimental intervention on body image, a 2 x 2 mixed between-within subjects ANOVA was conducted with condition (experimental vs control) as the between subjects factor and time (pre-manipulation vs post-manipulation) as the within subjects factor. All assumptions were upheld, including homogeneity of variance-covariance as indicated by Box’s M ( p >.001) and Levene’s ( p >.05) tests [ 64 ].
The interaction between condition and time was significant, Wilks’ Lambda = .98, F (1, 271) = 6.83, p = .009, partial eta squared = .03, demonstrating that the change in body image scores from pre-manipulation to post-manipulation was significantly different for the two groups. The body image satisfaction scores for women in both conditions decreased from pre-manipulation to post-manipulation. As anticipated, participants in the experimental condition reported a greater decrease in body image satisfaction than women in the control condition (see Table 4 ). This interaction effect is displayed in Fig 1 .
https://doi.org/10.1371/journal.pone.0307597.g001
https://doi.org/10.1371/journal.pone.0307597.t004
Although not consequential to the testing of the experimental manipulation, statistically significant main effects were also found for time, Wilks’ Lambda = .89, F (1, 271) = 32.99, p = < .001, partial eta squared = .109 and condition, F (1, 271) = 4.42, p = .036, partial eta squared = .016. The means and standard deviations of these main effects are displayed in Table 4 .
A second 2 x 2 mixed between-within subjects ANOVA was conducted to investigate the effect of the experimental manipulation on participants’ internalisation scores. All assumptions for the mixed model ANOVA were met with no violations.
A statistically significant interaction was found between group condition and time, Wilks’ Lambda = .97, F (1, 271) = 8.16, p = .005, partial eta squared = .029. This significant interaction highlights that the change in degree of internalisation at pre-manipulation and post-manipulation is not the same for the two conditions. Interestingly, the internalisation scores for women in the control group decreased from pre-manipulation to post-manipulation, whilst as anticipated, internalisation scores for women in the experimental group increased following exposure to the manipulation (see Table 5 ). This interaction is displayed in Fig 2 .
https://doi.org/10.1371/journal.pone.0307597.g002
https://doi.org/10.1371/journal.pone.0307597.t005
No statistically significant main effects were found for time, Wilks’ Lambda = .987, F (1, 271) = 3.59, p = .059, partial eta squared = .013 or condition, F (1, 271) = 2.65, p = .104, partial eta squared = .010. The means and standard deviations of internalisation scores for each condition at pre-manipulation and post-manipulation are displayed below in Table 5 .
The current study investigated the effect of TikTok content on women’s body image satisfaction and degree of internalisation of appearance ideals, and whether greater TikTok use contributed to increased disordered eating behaviour. In support of the hypotheses, exposure to pro-ana TikTok content significantly decreased participants’ body image satisfaction and increased participants’ degree of internalisation of appearance ideals. The hypothesis that greater daily TikTok use would contribute to increased disordered eating behaviour was not supported, as no statistically significant differences in restrictive disordered eating or ‘healthy’ disordered eating were found between the low, moderate, high, and extreme daily TikTok use groups.
Daily tiktok use and disordered eating behaviour..
Contrary to expectations, differences among groups on measures of restrictive disordered eating and ‘healthy’ disordered eating did not reach statistical significance. The proposed hypothesis that greater daily TikTok usage would be associated with disordered eating behaviour and attitudes was therefore unsupported. Despite lacking statistical support, participants categorised in the ‘high’ and ‘extreme’ daily TikTok use groups reported an average EAT-26 score of 18.16 and 19.09, respectively. Considering that an EAT-26 cut-off of ≥ 20 indicates potential clinical psychopathology, this mean score illustrates that exposure to TikTok content for two or more hours per day may contribute to a clinical degree of restrictive disordered eating.
The failure of the present study to detect any significant differences in disordered eating behaviours among participants with different TikTok daily usage does not align with the Transactional Model [ 33 ]. According to this model, risk factors such as low self-esteem and high thin ideal internalisation may predispose an individual to seek gratification via social media, resulting in body dissatisfaction and negative affect. The Transactional Model therefore proposes that a positive correlation exists between time spent on social media and body image dissatisfaction. Our findings also do not align with the conclusions Frieiro Padín et al. [ 34 ] drew from their review of the literature, in which a strong connection was identified between time on social media and heightened body image concerns and internalisation of the thin ideal, as well as eating disorder psychopathologies, though a distinction in outcome measures must be noted.
Based on the aforementioned sociocultural theory and previous research [see 28 , 43 , 48 ], it was assumed that increased body dissatisfaction as a result of increased time spent on social media (as stipulated by the Transactional Model), would lead to greater disordered eating behaviour. However, this was not supported statistically in the data. As postulated by Culbert et al. [ 69 ], disordered eating behaviour may instead only be a risk of media exposure if individuals are prone to endorse thin-ideals. Individuals in the present study that reported ‘high’ and ‘extreme’ daily TikTok use may have felt satisfied with their bodies and experienced lower thin-ideal internalisation. This could have potentially buffered the negative effect of greater TikTok content exposure and accounted for the lack of significant differences in disordered eating behaviour between groups. The quantity of TikTok consumption remains a pertinent question for disordered eating behaviour. As per the present study’s brief experimental manipulation, findings suggest that high frequency of daily TikTok use does not necessarily contribute to greater disordered eating behaviour than short exposures to this content.
Content presented to the pro-ana TikTok group included a mix of explicit and implicit pro- eating disorder messages as well as fitspiration content. Fitspiration content presented in the current study included workout videos to achieve a “smaller waist” and “toned abs” where female creators with slim, toned physiques sporting activewear took viewers through a series of exercises, advising viewers that they would “see results in a week”. In the present study, diet-related fitspiration content presented included the concoction of juices to “get rid of belly fat” and advice on the best “diet for a small waist” which requires avoidance of all meat, dairy, junk food, soda, and above all, to make “no excuses”. Fitspiration style content in the current study totalled one-minute, compared to disordered eating themes which totalled six minutes. The integration of these various types of content, although reflective of the For You function in TikTok, impeded our ability to determine the singular impact of fitspiration or disordered eating content, respectively, on body image and internalisation of societal beauty standards, but did reflect social media as it is consumed beyond experimental research settings.
Tiktok and body image states..
The hypothesis that women exposed to pro-ana TikTok content would experience a significant decrease in body image compared to women who viewed the control TikTok content was supported. The present study found a significant interaction effect of body image between group condition (control vs experimental) and time period (pre-manipulation vs post-manipulation), as well as significant main effects. It is important to note that the statistic of interest in evaluating the success of the experimental manipulation is the interaction effect, thus main effects must be interpreted secondarily and with caution [ 64 ]. Women in the experimental group reported significantly lower body image satisfaction after exposure to the pro-ana TikTok content and compared to women who viewed the control content. This finding corroborates Festinger’s [ 27 ] Social Comparison Theory that posits people naturally evaluate themselves in comparison to others. Exposure to the pro-ana TikTok content, consisting of various thin bodies and messaging around weight loss, may have provided the opportunity for women to engage in maladaptive upward social comparisons, resulting in reduced body image satisfaction. The present study upholds previous findings of Engeln-Maddox, Tiggemann, McComb and Mills, and Gibson [ 29 , 32 , 39 , 70 ] who suggest that visual exposure to thin bodies may adversely affect one’s level of body image satisfaction and extends this research by replicating this finding in the context of a contemporary media platform, TikTok, and by utilising an experimental design.
Contradicting the present study and previous research, Pryde and Prichard [ 42 ] found no significant increase in young women’s body dissatisfaction following exposure to fitspiration TikTok content. A potential explanation for this finding is that the performance of physical movements captured in fitspiration videos may shift the focus of viewers from aesthetics to functionality, highlighting physical competencies and capabilities which has been shown to improve body image satisfaction in young women [ 71 ]. Pryde and Prichard’s [ 42 ] fitspiration content did not include typically occurring harmful themes as the present study did, potentially reducing the negative implications for body image satisfaction of exposure to such content in real world contexts.
Interestingly, women in the control group also reported a statistically significant decrease in body image satisfaction after viewing the neutral TikTok content, a finding that underscores the possible complexity of social media’s influence on body image, as identified in research by Huülsing [ 72 ]. This is an unexpected finding, as the TikTok content displayed to the control group was selected specifically to be unrelated to appearance ideals and pressures. One possible reason for this result is the repetition of administration of the BISS within a short time period. Completing the BISS twice may have caused participants to focus more attention on their body appearance than usual, resulting in more critical appraisals regardless of the experimental stimuli to which they were exposed. This notion aligns with previous research that found focusing on the appearance of body was associated with lower body image satisfaction, whereas focus on the function of the body was associated with more positive body image states [ 71 ].
One potential explanation for this finding is that the control group stimuli was contaminated and produced an unintentional effect on body image scores. Two-minutes of footage within the seven-minute control group TikTok compilation presented the human body including legs, arms, and hands. Although this body-related content was neutral in nature, it may be that even ‘harmless’ representations of the human body are sufficient to elicit a social comparison response in participants or in some capacity, reinforce the #fitspiration motifs commonly depicted on TikTok [ 1 ], therefore impacting body image scores at post-manipulation. This possible explanation has implications for TikTok use and women’s body image, as it suggests that viewing even benign content of human bodies for less than 10-minutes can have an immediate detrimental impact on body image states, even when this content is unrelated to body dissatisfaction, thinness, or weight loss. Furthermore, although a statistically significant body image decrease was detected in the control group, this finding must be interpreted with caution due to the significant interaction effect obtained.
In accordance with the hypothesis, women in the experimental group reported a significant increase in their degree of internalisation of appearance ideals following exposure to pro-ana TikTok content. Women in the experimental group also reported significantly greater internalisation of appearance ideals than women in the control group. Conversely to the experimental group, internalisation scores of the control group decreased after viewing the neutral TikTok content. These findings are in line with the sociocultural theory, as women reported increased internalisation of societal beauty standards following exposure to media content explicitly and implicitly portraying the thinness ideal. The present study supports Mingoia et al’s. [ 53 ] meta-analysis, which yielded a positive association between social networking site use and the extent of internalisation of the thin ideal and furthers this notion by replicating the finding with TikTok specifically and utilising an experimental design.
In the current study, participants were subject to a single brief exposure of pro-ana TikTok content, whereas most of the sample indicated that their TikTok use was up to two hours per day. This suggests that the degree of internalisation of appearance ideals in participants lives outside of the experiment are likely to be much greater. Mingoia et al. [ 53 ] also found that the use of appearance-related features on social networking sites, such as posting and viewing photos and videos, demonstrated a stronger relationship with the internalisation of the thin ideal than the use of social networking features that were not appearance-related, such as messaging and writing status updates. As TikTok is a video sharing app and most of its content generally features full-body-length camera shots rather than a face or head shot, this finding suggests that TikTok users could potentially internalise body-related societal standards to a greater extent than users of other social media apps that typically feature head shots.
The finding that women internalised societal beauty standards to a greater degree after being exposed to pro-ana TikTok content corroborates the sociocultural theory’s emphasis of the significance of social influences in internalisation. TikTok users may be exposed to all three social influences (i.e., media, peers, and family) simultaneously on a single platform which may encourage internalisation of appearance-ideals in a more profound manner than any of these three influences in isolation. One point of difference between TikTok and other social media apps is that much content on the app is generated by “ordinary” individuals, rather than supermodels or celebrities. This enables blatantly insidious and diet-related content to circulate the app with less policing and scrutiny compared to content produced by an influencer or celebrity who may be more likely to be criticised or cancelled for socially irresponsible messaging and also provides the opportunity for more horizontal social comparisons and peer-to-peer style interactions rather than upward social comparisons.
Indeed, in their study of American teens, Mueller et al. [ 52 ] identified that girls were especially likely to engage in weight loss behaviour if a high proportion of girls with a similar BMI were also engaging in weight loss behaviours. This indicates that internalisation was strongest when appearance-ideals were promoted by alike peers. Due to the fact that much pro-ana TikTok content is created by young women, Mueller et al’s. [ 52 ] finding has problematic implications for the young female users of TikTok, in that harmful diet-related messages could be internalised to a greater extent on TikTok than on other platforms and potentially lead to body image disturbances, disordered eating behaviour, and other negative outcomes among young women.
The findings of the current study are important but must also be understood within the broader context of participant’s daily lives beyond their participation in this study. Everyday female-identifying individuals are exposed to a multitude of different sources of information from which body image related stimuli can be drawn. The present study’s experiment was not conducted in a controlled environment due to its online nature, therefore researchers did not have the ability to assess and control for other pieces of body image-related information that participants might have consumed prior to participation that may have been salient for their body image. Further research is required to identify how sustained a change in body image states as measured by the BISS may be over time.
The findings of this study provide some insights into how social media influences disordered eating behaviour and mental health; a theoretical gap in the literature that Choukas-Bradley et al. [ 6 ] highlight as holding back research in this domain. In particular, the findings of the current study indicate that short periods of exposure to disordered TikTok content have an effect, while the high-range EAT-26 scores observed for those who engaged with TikTok for two or more hours a day also raise questions about the duration of exposure. Nonetheless, our findings demonstrate that short exposure periods are sufficient to have a negative effect on body image and internalisation of the thin ideal.
One point that may be readily overlooked in developing a theoretical framework around social media’s influence is that the narrative arc of TikTok videos is such that users are exposed to many short stories in quick succession, which may have a different effect to longer form content from a single content creator. As Pierce [ 2 ] notes, the speed of exposure to overlapping, but separate narratives depicted in successive videos, is an important feature of TikTok content and may contribute to the influence of such platforms on disordered eating and body attitudes. Each piece of content serves as a standalone narrative but may also overlap and interact with the viewer’s experience of the next video they watch to build a cumulative, normalised narrative of disordered body- and eating-practices.
In the current study, participants who engaged with TikTok for two-three hours a day were classified as high users, and those who used TikTok for three or more hours were classed as extreme. These rates of usage may, however, be quite normative, with Santarossa and Woodruff [ 73 ] citing three-four hours a day on social media as normative for their sample of young adults, though notably participants in the current study were only questioned about their TikTok usage, not their general use of social media.
While we examined the effect of pro-ana content in this study, that some changes were observed in the control group as well as the experimental group indicates that the social media environment, characterised as it is by idealisation, instant feedback, and readily available social comparison [ 6 ], may play a general role in diminishing positive body image attitudes and healthy aspirations. This is supported by Tiggemann and Slater’s [ 35 , 36 ] research in which social media usage was found to correlate positively with higher levels of body image concerns, in contrast to time spent on the internet more generally, and this may be particularly true for visually oriented platforms that sensitize viewers to their own appearance and that of others. As noted previously, of the visually-oriented social media platforms, predominantly TikTok and Instagram, videos are commonly framed on TikTok so that the subject’s whole body is visible, particularly in dance videos and in #GymTok content, where on Instagram, cameo style head-shot videos appear more likely to feature, which further suggests that TikTok may provide more body-related stimuli than other platforms, even when the intention of the content does not relate to body-image or #fitspiration.
Importantly, the algorithm on TikTok functions in such a way that those who actively seek out body positivity content may also be exposed to nefarious body-related content such as body checking, a competitive, self-surveillance type of content where users are encouraged to test out their weight by attempting to drink from a glass of water while their arm encircles another’s waist. As McGuigan [ 74 ] reports, watching just one body checking video may result in hundreds more filtering through a user’s For You page, with those actively attempting to seek out positive body image content likely to be inadvertently exposed to disordered content due to the configuration of the algorithm. This function of the For You page is demonstrated in the current study, with 64% of participants reporting having seen disordered eating content on their For You page, higher than any other kind of harmful content, including suicide and bullying. The current study did not assess participants’ consumption of #FoodTok, #GymTok, and #Fitspiration. Engagement with these dimensions of TikTok and the type of content that participants seek out via the search function warrant consideration in future research.
The TikTok algorithm underscores Logrieco et al’s. [ 18 ] findings that even anti-anorexia content can be problematic, especially given complexities in determining and controlling what is performatively problematic, including videos discussing recovery and positive body attitudes that may somewhat paradoxically further body policing and competition among users and consumers of social media content. Furthermore, as Logrieco et al. [ 18 ] highlight, TikTok is replete in both pro-ana and much more implicit body-related content that may be harmful to viewers, not to mention those creating the content, whose experiences also warrant consideration.
The present study bridged an important gap in the literature by utilising both experimental and cross-sectional designs to examine the influence of pro-ana TikTok content on users’ body image satisfaction, internalisation of body ideals, and disordered eating behaviours. While the negative impact of social media on body image and eating behaviours has been established in relation to platforms such as Instagram and Twitter, TikTok’s rapid emergence and unique algorithm warrant independent analysis.
The present findings have important theoretical implications for the understanding of sociocultural influences of orthorexia nervosa development. Notably, this study is one of the first to highlight the association between orthorexia nervosa and the tripartite model of disordered eating using an experimental design. The results illustrate that the internalisation of sociocultural appearance ideals predicts the development of ‘healthy’ disordered eating, as suggested by the tripartite theory. Western culture ideals do seem to influence the expression of orthorexic tendencies, thus caution should be exercised by women when interacting with appearance-related TikTok content.
Unlike explicit pro-ana content, which is open to condemnation, the moral and health-related discourses underpinning much body-related content in which thinness and health are espoused as goodness, reflects a new trend in diet culture masquerading as wellness culture [ 20 , 21 ]. Questions are raised around the ethics of social media algorithms when the technologically fostered link between recovery-focused content and disordered-content on TikTok is laid bare, particularly considering that extant research has found individuals with experience of eating disorders often seek out support, safety, and connection online [ 49 ] and in doing so on a platform like TikTok, may be exposed to more disordered eating content than the average user. Given visual social media platforms are associated with higher levels of dysfunction in relation to body image [ 4 ], the policy and ethics of such platforms warrant scrutiny from a variety of stakeholders in management, marketing, technology regulation, with psychology playing an important role in the marketing of these platforms. As traditional journalistic platforms have been subjected to scrutiny and reform, so too must a climate of accountability be established within the social media nexus.
The widespread growth of social media may warrant greater concern than traditional forms of mass media, not only because of the full-time accessibility and diverse range of platforms, but also due to the prevalence of peer-to-peer interactions. According to the social comparison theory, comparison of oneself to others has traditionally considered more removed, higher status influences (e.g., celebrities, actors/actresses, supermodels) as a greater source of pressure than those in the individuals’ natural environment (e.g., family and peers). Re-examination of this theoretical perspective is warranted considering the contemporary challenges of social media and the perpetuation of body image messages from alike peers. Furthermore, a diverse range of “content” may trigger disordered body- and eating-related attitudes, including #fitspiration and #GymTok, which poses challenges for social media platforms in regulating content. The inclusion of orthorexia in the milieu highlights the disordered nature of seemingly benign health practices and social media content.
That TikTok content containing explicit and implicit pro-ana themes may readily remain on the app uncensored exemplifies the importance of protective strategies to build resilience at the individual level. One such protective strategy is shifting focus from body appearance to functionality. Alleva and colleagues [ 71 ] investigated the Expand Your Horizon programme, designed to improve body image by training women to focus on body functionality. They report that women who engaged with the Expand Your Horizon programme experienced greater satisfaction with body image and functionality, body appreciation, and reduced self-objectification compared to women who did not engage with the program. Health professionals involved in the care of women with eating disorders and other mental health issues should also be educated to ensure they are knowledgeable about the social media content their clients may be exposed to, equipping them with skills to engage in conversations about the potential detrimental impacts of viewing pro-ana and other harmful TikTok content [ 53 ].
The administration of such programs in schools, universities, community groups, and clinical settings could prove effectual in the prevention of disordered eating and body image disturbance development and may reduce symptom severity of a pre-established disorder. Such programs must be developed with great care, however, given the propensity for even anti-anorexia content to have a negative effect on those consuming it [ 18 ]. The development of self-compassion may also build resilience in women, with research confirming that self-compassion can be effectively taught [ 75 ]. Subsequently, programs have been developed such as Compassion Focused Therapy (CFT) in which clients are trained to develop more compassionate self-talk during negative thought processes and to foster more constructive thought patterns [ 76 ]. The value of CFT has been established in the literature with both clinical and non-clinical samples and has promising outcomes particularly for those high in self-criticism [ 77 ].
Young women should be provided with media literacy tools that can assist in advancing critical evaluations of the online world. Digital manipulation of advertising and celebrity images is well known to many people, however, this awareness may be lacking regarding social media images, as they are generally disseminated within one’s peer network rather than outside of it [ 33 ]. Media literacy interventions may educate women about how social media perpetuates appearance-ideals that are often unrealistic and unattainable [ 53 ]. As an example, Posavac et al. [ 78 ] revealed that a single media literacy intervention resulted in a reduction in women’s social comparison to body ideals portrayed in the media.
Such interventions might be extended to female-identifying TikTok users to educate them on the manipulation of videos to produce idealised portrayals of the self. Media literacy should be commenced from an early age by teaching children, adolescents, and adults to understand the influence of implicit messages conveyed through social media and to create media content that is responsible and psychologically safe for others [ 79 ]. Increased understanding of messages portrayed by social media content may prevent thin-ideal endorsement and internet misuse. Notably, however, the most effective approach would be to address the problem at its source and increase the regulation of social media companies, rather than upskill users in how to respond to harmful online environments, which creates further labour for the individual while allowing organisations to continue to produce harmful but easily monetizable content.
To meet the requirements to run multivariate analyses, the continuous data of body image and internalisation scores were dichotomised using a median split to create ‘low’ and ‘high’ groups for each variable. Although dichotomisation was necessary to perform appropriate analyses and power analyses deemed the sample size as adequate following performance of the median split, dichotomising these variables may have contributed to a loss of statistical power to detect true effects.
Limitations are implicated in the use of the ORTO-15 in the present study. The ORTO-15 does not account for different lifestyle factors that may alter a participants’ response, such as dietary restrictions, food intolerances, or medical dietary guidelines. The discrepancies in literature surrounding the psychometric properties of the ORTO-15 may be attributable to the lack of established diagnostic criteria of orthorexia nervosa, cultural differences in expressions of eating disorders, and difficulty comparing research results in determining orthorexia nervosa diagnoses due to inconsistencies in testing questions and cut-off values [ 61 ]. Due to unacceptable reliability in the present study, a factor analysis was performed which identified a factor relating to health food preoccupation. This identified factor was used as the ORTO-15 measure and data from these 5-items were used in analyses and referred to throughout the present study as ‘healthy’ disordered eating. Using the 5-items related to ‘healthy’ disordered eating rather than the complete 15-item scale may not have accurately assessed participants’ degree of orthorexic tendencies. Despite these limitations, the ORTO-15 is the only accepted measure of orthorexic tendencies available [ 63 ]. Additionally, more limitations would likely have been encountered by using the full 15-item measure lacking reliability, compared to utilising the 5-item factor with acceptable reliability.
Future studies of TikTok and disordered eating behaviour should incorporate a measure of social comparison to verify whether social comparison is the vehicle through which women experience decreased body image satisfaction after viewing TikTok content. Future research should also examine the influence of TikTok content creation on body image, internalisation of thinness, and disordered eating behaviour and explore the association between what individuals consume on TikTok and the social media content that they produce. This research should be conducted using more diverse samples of women, including transgender women, to determine whether the findings of the present study are relevant for this population given the unique challenges regarding body image and societal beauty standards that they may experience.
Longitudinal studies are also warranted to examine the effect of exposure to pro-ana TikTok content over time, and to assess the effects of pro-ana TikTok content on body image satisfaction and eating disorder symptomology over time. Further research on orthorexia nervosa is needed to establish a more reliable measure of orthorexic tendencies and this would enable future investigation of the impact of pro-ana TikTok content on the development of orthorexia nervosa, as well as individual differences as predisposing factors in the development of orthorexic tendencies. Finally, future research should examine the efficacy of media literacy and self-compassion intervention programs as a protective factor specifically in the TikTok context, where disordered eating messages are more explicit in nature than traditional media and other social media platforms.
The findings of the current study support the notion that pro-ana TikTok content decreases body image satisfaction and increases internalisation of societal beauty standards in young women. This research is timely given reliance on social media for social interaction, particularly for young adults. Our findings indicate that female-identifying TikTok users may experience psychological harm even when explicit pro-ana content is not sought out and even when their TikTok use is time-limited in nature. The findings of this study suggest cultural and organisation change is needed. There is a need for more stringent controls and regulations from TikTok in relation to pro-ana content as well as more subtle forms of disordered eating- and body-related content. Prohibiting or restricting access to pro-ana content on TikTok may reduce the development of disordered eating and the longevity and severity of established eating disorder symptomatology among young women in the TikTok community. There are current steps being taken to delete dangerous content, including blocking searches such as “#anorexia”, however, there are various ways users circumvent these controls and further regulation is required. Unless effective controls are implemented within the platform to prevent the circulation of pro-ana content, female-identifying TikTok users may continue to experience immediate detrimental consequences for body image satisfaction, thin-ideal internalisation, and may experience an increased risk of developing disordered eating behaviours.
Breaking , more commonly known as breakdancing, made its debut as an Olympic sport this week at the 2024 Paris Games , with 17 B-girls and 16 B-boys making their way to France with the hopes of securing a gold medal.
On the first day of competition, viewers from across the world were treated to a different kind of introduction — not to the sport itself, but one of its athletes.
Though she was a long way from winning a gold medal, likely no breaker Friday captured the imagination of the international audience more than Rachael Gunn, an Australian breaker who competes under the name “Raygun.”
REQUIRED READING: Follow USA TODAY's coverage of the 2024 Paris Olympics
Raygun went 0-3 in her head-to-head competitions Friday — falling to Logistx of the United States, Syssy of France and eventual silver medalist Nicka of Lithuania by a combined score of 54-0 — and failed to record a point across those three matches, but for what she lacked in smoothly executed moves, she made up for in the hearts she won over with her demeanor.
Raygun’s short-lived Olympic experience made her a celebrity, one who people became even more enamored with once they learned more about her.
The 36-year-old Gunn, who was one of the oldest qualifiers in the breaking competition, has a PhD in cultural studies and is a college professor at Macquarie University in Sydney. Her research focuses primarily on breaking, street dance and hip-hop culture while her work draws on “cultural theory, dance studies, popular music studies, media, and ethnography.”
“In 2023, many of my students didn’t believe me when I told them I was training to qualify for the Olympics, and were shocked when they checked Google and saw that I qualified,” Gunn said to CNBC earlier this month .
Unlike much of her competition in Paris, Gunn took up break dancing later in life. She didn’t enter her first battle until 2012.
On Friday, a person who began the day as a little-known academic ended it as a viral worldwide sensation.
Here’s a sampling of the reaction to Raygun and her performance:
2024 PARIS OLYMPICS: Meet the members of Team USA competing at the 2024 Paris Olympics
I could live all my life and never come up with anything as funny as Raygun, the 36-year-old Australian Olympic breakdancer pic.twitter.com/1uPYBxIlh8 — mariah (@mariahkreutter) August 9, 2024
Give Raygun the gold right now #breakdancing pic.twitter.com/bMtAWEh3xo — n★ (@nichstarr) August 9, 2024
my five year old niece after she says “watch this!” : pic.twitter.com/KBAMSkgltj — alex (@alex_abads) August 9, 2024
I'd like to personally thank Raygun for making millions of people worldwide think "huh, maybe I can make the Olympics too" pic.twitter.com/p5QlUbkL2w — Bradford Pearson (@BradfordPearson) August 9, 2024
The Aussie B-Girl Raygun dressed as a school PE teach complete with cap while everyone else is dressed in funky breaking outfits has sent me. It looks like she’s giving her detention for inappropriate dress at school 🤣 #Olympics pic.twitter.com/lWVU3myu6C — Georgie Heath🎙️ (@GeorgieHeath27) August 9, 2024
There has not been an Olympic performance this dominant since Usain Bolt’s 100m sprint at Beijing in 2008. Honestly, the moment Raygun broke out her Kangaroo move this competition was over! Give her the #breakdancing gold 🥇 pic.twitter.com/6q8qAft1BX — Trapper Haskins (@TrapperHaskins) August 9, 2024
my dog on the lawn 30 seconds after i've finished bathing him pic.twitter.com/A5aqxIbV3H — David Mack (@davidmackau) August 9, 2024
My wife at 3AM: I think I heard one of the kids Me: No way, they are asleep *looks at baby monitor* pic.twitter.com/Ubhi6kY4w4 — Wes Blankenship (@Wes_nship) August 9, 2024
me tryna get the duvet off when i’m too hot at night #olympics pic.twitter.com/NM4Fb2MEmX — robyn (@robynjournalist) August 9, 2024
Raygun really hit them with the "Tyrannosaurus." pic.twitter.com/ZGCMjhzth9 — Mike Beauvais (@MikeBeauvais) August 9, 2024
Raygun (AUS) https://t.co/w2lxLRaW2x — Peter Nygaard (@RetepAdam) August 9, 2024
IMAGES
COMMENTS
Step 1: Define your variables. You should begin with a specific research question. We will work with two research question examples, one from health sciences and one from ecology: Example question 1: Phone use and sleep. You want to know how phone use before bedtime affects sleep patterns.
When to use Experimental Research Design . Experimental research design should be used when a researcher wants to establish a cause-and-effect relationship between variables. It is particularly useful when studying the impact of an intervention or treatment on a particular outcome. Here are some situations where experimental research design may ...
Experimental research is a study conducted with a scientific approach using two sets of variables. The first set acts as a constant, which you use to measure the differences of the second set. Quantitative research methods, for example, are experimental.
Using strictly controlled environments, behaviorists were able to isolate a single stimulus as the cause of measurable differences in behavior or physiological responses. The foundations of social learning theory and behavior modification are found in experimental research projects.
The pre-experimental research design is further divided into three types. One-shot Case Study Research Design. In this type of experimental study, only one dependent group or variable is considered. The study is carried out after some treatment which was presumed to cause change, making it a posttest study.
10 Experimental research. 10. Experimental research. Experimental research—often considered to be the 'gold standard' in research designs—is one of the most rigorous of all research designs. In this design, one or more independent variables are manipulated by the researcher (as treatments), subjects are randomly assigned to different ...
Experimental research design is a framework of protocols and procedures created to conduct experimental research with a scientific approach using two sets of variables. Herein, the first set of variables acts as a constant, used to measure the differences of the second set. The best example of experimental research methods is quantitative research.
Quasi experiments do not use randomization to assign research subjects to experimental conditions; instead, some other method of assignment is utilized. Often research subjects voluntarily choose to participate or not to participate in the treatment of interest. Thus, the actions and wishes of the research subjects typically affect assignment.
Experimental research serves as a fundamental scientific method aimed at unraveling. cause-and-effect relationships between variables across various disciplines. This. paper delineates the key ...
Step 1: Define your variables. You should begin with a specific research question. We will work with two research question examples, one from health sciences and one from ecology: Example question 1: Phone use and sleep. You want to know how phone use before bedtime affects sleep patterns.
Experimental research design is a scientific framework that allows you to manipulate one or more variables while controlling the test environment. When testing a theory or new product, it can be helpful to have a certain level of control and manipulate variables to discover different outcomes. You can use these experiments to determine cause ...
Experimental research is commonly used in sciences such as sociology and psychology, physics, chemistry, biology and medicine etc. It is a collection of research designs which use manipulation and controlled testing to understand causal processes. Generally, one or more variables are manipulated to determine their effect on a dependent variable.
The three main types of experimental research design are: 1. Pre-experimental research. A pre-experimental research study is an observational approach to performing an experiment. It's the most basic style of experimental research. Free experimental research can occur in one of these design structures: One-shot case study research design: In ...
Experimental science is the queen of sciences and the goal of all speculation. Roger Bacon (1214-1294) Experiments are part of the scientific method that helps to decide the fate of two or more competing hypotheses or explanations on a phenomenon. The term 'experiment' arises from Latin, Experiri, which means, 'to try'.
The terms study design, experimental design, and research design are often thought to be synonymous and are sometimes used interchangeably in a single paper. Avoid doing so. Use the term that is preferred by the style manual of the journal for which you are writing. Study design is the preferred term in the AMA Manual of Style, 2 so I will use ...
The experimental method involves manipulating one variable to determine if this causes changes in another variable. This method relies on controlled research methods and random assignment of study subjects to test a hypothesis. For example, researchers may want to learn how different visual patterns may impact our perception.
Here are some examples of experimental research: This research method can be used to evaluate employees' skills. Organizations ask candidates to take tests before filling a post. It is used to screen qualified candidates from a pool of applicants. This allows organizations to identify skills at the time of employment.
In this case, quasi-experimental research involves using intact groups in an experiment, rather than assigning individuals at random to research conditions. (some researchers define this latter situation differently. For our course, we will allow this definition). In causal comparative (ex post facto) research, the groups are already formed. It ...
An example of experimental research in marketing: The ideal goal of a marketing product, advertisement, or campaign is to attract attention and create positive emotions in the target audience. Marketers can focus on different elements in different campaigns, change the packaging/outline, and have a different approach.
Collect the data by using suitable data collection according to your experiment's requirement, such as observations, case studies , surveys , interviews, questionnaires, etc. Analyse the obtained information. Step 8. Present and Conclude the Findings of the Study. Write the report of your research.
An experimental research design is typically focused on the relationship between two variables: the independent variable and the dependent variable. The researcher uses random sampling and random ...
There are three types of experiments you need to know: 1. Lab Experiment. A laboratory experiment in psychology is a research method in which the experimenter manipulates one or more independent variables and measures the effects on the dependent variable under controlled conditions. A laboratory experiment is conducted under highly controlled ...
We provide new evidence on the causal effect of unearned income on consumption, balance sheets, and financial outcomes by exploiting an experiment that randomly assigned 1000 individuals to receive $1000 per month and 2000 individuals to receive $50 per month for three years. The transfer increased ...
Dong Q, Xu G, Chen W. Experimental Research on the Low-Cycle Fatigue Crack Growth Rate for a Stiffened Plate of EH36 Steel for Use in Ship Structures. Journal of Marine Science and Engineering . 2024; 12(8):1365.
Artificial Intelligence. NIH promotes the safe and responsible use of AI in biomedical research through programs that support the development and use of algorithms and models for research, contribute to AI-ready datasets that accelerate discovery, and encourage multi-disciplinary partnerships that drive transparency, privacy, and equity.
6 Because we are interested in the potential for using satisficing as a screener in research and sought to directly compare it to the three sample sources, we wanted to examine groups of participants that would yield three groups, similar to our three sample sources, as shown in Table 3.However, we also performed a Model 3 analysis in PROCESS using the quantitative measure of response satisficing.
To address growing concerns of tire pollution and a specific pollutant called 6PPD-quinone (6PPD-Q), EPA researcher Dr. Paul Mayer led an effort to investigate the life cycle of tires and their impacts on the environment.The resulting article, "Where the rubber meets the road: Emerging environmental impacts of tire wear particles and their chemical cocktails," is a holistic examination and ...
Based on the experimental results, a suite of models has been developed to characterize the accuracy of CRAM computation. ... Research aiming at connecting logic and memory more closely has led to ...
Videos glamourising disordered eating practices and body image concerns readily circulate on TikTok. Minimal empirical research has investigated the impact of TikTok content on body image and eating behaviour. The present study aimed to fill this gap in current research by examining the influence of pro-anorexia TikTok content on young women's body image and degree of internalisation of ...
Breaking, more commonly known as breakdancing, made its debut as an Olympic sport this week at the 2024 Paris Games, with 17 B-girls and 16 B-boys making their way to France with the hopes of ...