Georgia Institute of Technology
Cryogenic Compute-In-Memory Circuits and Architectures for Spiking Neural Networks
Abstract
dc:description.abstractThis work presents a Computing-In-Memory (CIM)-based Spiking Neural Network (SNN) architecture for energy-efficient, low-latency inference in cryogenic conditions. The proposed CIM-SNN integrates current-domain 8T SRAM vector-matrix multiplication, cryogenic-aware temporal encoding, and time-to-first-spike (TTFS) output readout in a unified signal path with low-voltage operation at 77 K. Using calibrated 14 nm bulk FinFET models, the design exploits cryogenic device behavior, including reduced static leakage, steeper subthreshold switching, and enhanced mobility/transconductance, to design the SRAM array interface, neuron circuitry, and inter-layer temporal encoder. A Cryo-Integrated Spike Encoder (CISE) is proposed to replace the conventional multi-stage neuron and temporal-encoding interface with a current-to-time conversion block, reducing conversion overhead between current-domain SRAM computation and temporal inter-layer communication. For MNIST classification, the optimized cryogenic design achieves a 4.54× improvement in energy-delay product (EDP) over room-temperature operation while maintaining similar classification accuracy. Further, this work designs an optimized CIM-SNN architecture for autonomous health monitoring of miniaturized satellites at cryogenic temperature, consuming less than 1 mW of power.
Degree
thesis:*- Name thesis:degree_name
- Electrical and Computer Engineering, MS
- Department thesis:degree_department
- Electrical and Computer Engineering
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Shamieh, Laith A.
- Advisor dc:contributor.advisor
-
- Mukhopadhyay, Saibal
- Committee members dc:contributor.committeemember
-
- Datta, Suman
- Sathe, Visvesh
- ORCID person:identifier.orcid
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81659