MINIMALIST: switched-capacitor circuits for efficient in-memory computation of gated recurrent units

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Billaudelle, Sebastian, Kriener, Laura, Moro, Filippo, Torchet, Tristan, Payvand, Melika
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913834646634496
author Billaudelle, Sebastian
Kriener, Laura
Moro, Filippo
Torchet, Tristan
Payvand, Melika
author_facet Billaudelle, Sebastian
Kriener, Laura
Moro, Filippo
Torchet, Tristan
Payvand, Melika
contents Recurrent neural networks (RNNs) have been a long-standing candidate for processing of temporal sequence data, especially in memory-constrained systems that one may find in embedded edge computing environments. Recent advances in training paradigms have now inspired new generations of efficient RNNs. We introduce a streamlined and hardware-compatible architecture based on minimal gated recurrent units (GRUs), and an accompanying efficient mixed-signal hardware implementation of the model. The proposed design leverages switched-capacitor circuits not only for in-memory computation (IMC), but also for the gated state updates. The mixed-signal cores rely solely on commodity circuits consisting of metal capacitors, transmission gates, and a clocked comparator, thus greatly facilitating scaling and transfer to other technology nodes. We benchmark the performance of our architecture on time series data, introducing all constraints required for a direct mapping to the hardware system. The direct compatibility is verified in mixed-signal simulations, reproducing data recorded from the software-only network model.
format Preprint
id arxiv_https___arxiv_org_abs_2505_08599
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MINIMALIST: switched-capacitor circuits for efficient in-memory computation of gated recurrent units
Billaudelle, Sebastian
Kriener, Laura
Moro, Filippo
Torchet, Tristan
Payvand, Melika
Hardware Architecture
Artificial Intelligence
Machine Learning
Signal Processing
Recurrent neural networks (RNNs) have been a long-standing candidate for processing of temporal sequence data, especially in memory-constrained systems that one may find in embedded edge computing environments. Recent advances in training paradigms have now inspired new generations of efficient RNNs. We introduce a streamlined and hardware-compatible architecture based on minimal gated recurrent units (GRUs), and an accompanying efficient mixed-signal hardware implementation of the model. The proposed design leverages switched-capacitor circuits not only for in-memory computation (IMC), but also for the gated state updates. The mixed-signal cores rely solely on commodity circuits consisting of metal capacitors, transmission gates, and a clocked comparator, thus greatly facilitating scaling and transfer to other technology nodes. We benchmark the performance of our architecture on time series data, introducing all constraints required for a direct mapping to the hardware system. The direct compatibility is verified in mixed-signal simulations, reproducing data recorded from the software-only network model.
title MINIMALIST: switched-capacitor circuits for efficient in-memory computation of gated recurrent units
topic Hardware Architecture
Artificial Intelligence
Machine Learning
Signal Processing
url https://arxiv.org/abs/2505.08599