Prospects of analog in-memory computing using ferroelectric tunnel junctions
(2025) In Neuromorphic Computing and Engineering 5(2).- Abstract
Artificial intelligence (AI) is set to disrupt the way businesses and civil society operate, but the large energy usage for training and running AI compute models remains a pressing concern in terms of its sustainability. Analog in-memory computing (AIMC) hardware based on memristor devices is a promising route to significant energy-savings. Among the various memristor technologies, ferroelectric tunnel junctions (FTJs) hold considerable promise for large scale AIMC. Their performance prospects remain to be fully assessed; therefore, we here evaluate the viability of FTJ memristors as compute units for AIMC. Based on the behavior of experimental TiN/HfZrO4/W FTJs we define three operating modes which we apply to standard AI... (More)
Artificial intelligence (AI) is set to disrupt the way businesses and civil society operate, but the large energy usage for training and running AI compute models remains a pressing concern in terms of its sustainability. Analog in-memory computing (AIMC) hardware based on memristor devices is a promising route to significant energy-savings. Among the various memristor technologies, ferroelectric tunnel junctions (FTJs) hold considerable promise for large scale AIMC. Their performance prospects remain to be fully assessed; therefore, we here evaluate the viability of FTJ memristors as compute units for AIMC. Based on the behavior of experimental TiN/HfZrO4/W FTJs we define three operating modes which we apply to standard AI tests and two real-world use cases: Image segmentation (YOLOv8) and natural language processing (BERT). We find that the inherent mechanism behind the analog state in FTJs limits the usable dynamic range and thus enforces strict control of noise. The best overall performance is obtained by constricting the dynamic range further and using ultrashort programming pulses (0.5 ns); on BERT matching the performance of digital hardware. We also present a more accessible approach, combining three FTJs, two of which are operated as binary memories, to achieve similar performance while also relaxing the requirements on data converter precision and level of noise. All in all, we here reveal benefits and intrinsic challenges of FTJ-based AIMC systems, providing a blueprint for future experimental implementations.
(Less)
- author
- Borg, Mattias
LU
; Papadopoulos, Christos ; Guerin, Alec ; Athle, Robin LU and Bastani, Saeed LU
- organization
- publishing date
- 2025-06-01
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- analog in-memory computing (AIMC), BERT, ferroelectric tunnel junction (FTJ), hafnium zirconate (HZO), memristor, neuromorphic computing, Yolo
- in
- Neuromorphic Computing and Engineering
- volume
- 5
- issue
- 2
- article number
- 024006
- publisher
- IOP Publishing
- external identifiers
-
- scopus:105005143211
- ISSN
- 2634-4386
- DOI
- 10.1088/2634-4386/add0d9
- project
- Robust neuromorphic computing using ferroelectric memristors
- Ultra-fast thermal processing for next-generation ferroelectric hafnia
- language
- English
- LU publication?
- yes
- additional info
- Publisher Copyright: © 2025 The Author(s). Published by IOP Publishing Ltd.
- id
- 17fc1c09-b564-480d-9e1f-dbd55a93784a
- date added to LUP
- 2025-06-09 11:30:00
- date last changed
- 2025-06-10 12:19:29
@article{17fc1c09-b564-480d-9e1f-dbd55a93784a, abstract = {{<p>Artificial intelligence (AI) is set to disrupt the way businesses and civil society operate, but the large energy usage for training and running AI compute models remains a pressing concern in terms of its sustainability. Analog in-memory computing (AIMC) hardware based on memristor devices is a promising route to significant energy-savings. Among the various memristor technologies, ferroelectric tunnel junctions (FTJs) hold considerable promise for large scale AIMC. Their performance prospects remain to be fully assessed; therefore, we here evaluate the viability of FTJ memristors as compute units for AIMC. Based on the behavior of experimental TiN/HfZrO<sub>4</sub>/W FTJs we define three operating modes which we apply to standard AI tests and two real-world use cases: Image segmentation (YOLOv8) and natural language processing (BERT). We find that the inherent mechanism behind the analog state in FTJs limits the usable dynamic range and thus enforces strict control of noise. The best overall performance is obtained by constricting the dynamic range further and using ultrashort programming pulses (0.5 ns); on BERT matching the performance of digital hardware. We also present a more accessible approach, combining three FTJs, two of which are operated as binary memories, to achieve similar performance while also relaxing the requirements on data converter precision and level of noise. All in all, we here reveal benefits and intrinsic challenges of FTJ-based AIMC systems, providing a blueprint for future experimental implementations.</p>}}, author = {{Borg, Mattias and Papadopoulos, Christos and Guerin, Alec and Athle, Robin and Bastani, Saeed}}, issn = {{2634-4386}}, keywords = {{analog in-memory computing (AIMC); BERT; ferroelectric tunnel junction (FTJ); hafnium zirconate (HZO); memristor; neuromorphic computing; Yolo}}, language = {{eng}}, month = {{06}}, number = {{2}}, publisher = {{IOP Publishing}}, series = {{Neuromorphic Computing and Engineering}}, title = {{Prospects of analog in-memory computing using ferroelectric tunnel junctions}}, url = {{http://dx.doi.org/10.1088/2634-4386/add0d9}}, doi = {{10.1088/2634-4386/add0d9}}, volume = {{5}}, year = {{2025}}, }