Contribution to a conference proceedings

FZJ-2024-00724

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Non-idealities and Design Solutions for Analog Memristor-Based Content-Addressable Memories

Manea, P. (Corresponding author)FZJ* ; Sudarshan, C.FZJ* ; Cüppers, F.FZJ* ; Strachan, J. P.FZJ*

2023

Abstract: Memristor-based analog Content Addressable Memories (aCAMs)offer robust parallel pattern look-up capabilities, significantly en-hancing the scope of In-Memory Computing applications. Thispaper presents challenges of these analog circuits, which may oc-cur during the inference, and proposes solutions to overcome them.Precisely, we investigate the impact of temperature-dependent be-havior, CMOS process variations and memristor telegraph readnoise. We demonstrate that one challenging issue affecting memris-tors analog computing applications, namely telegraph read noise, isnot a significant problem in aCAM. We introduce a framework thataccounts for these combined distortions to define variability-awareaCAM windows and estimate the bit resolution of a CAM cell. Usingthis framework we estimate the bit resolution to 2 bits before apply-ing compensating measures and to 4 bits afterwards. We study howvariations affect the inference accuracy of the IRIS classificationdataset using our novel torchCAM model. We introduce a stream-lined aCAM design featuring a memristor comparator for simplifiedinput-to-reference comparison and a novel cell architecture withtwo symmetrical memristor comparator units.

---

Contributing Institute(s):

1. Neuromorphic Compute Nodes (PGI-14)

Research Program(s):

1. 5234 - Emerging NC Architectures (POF4-523) (POF4-523)
2. 5233 - Memristive Materials and Devices (POF4-523) (POF4-523)
3. 5232 - Computational Principles (POF4-523) (POF4-523)
4. BMBF 16ME0398K - Verbundprojekt: Neuro-inspirierte Technologien der künstlichen Intelligenz für die Elektronik der Zukunft - NEUROTEC II - (BMBF-16ME0398K) (BMBF-16ME0398K)

Appears in the scientific report 2024

---