S. N. B. Tushar, S. H. Alam, N. N Chakraborty and C. D. Schuman

April, 2026

IEEE 19th Dallas Circuits and Systems Conference (DCAS)

https://ieeexplore.ieee.org/abstract/document/11544440

PDF not available yet, or is only available from the conference/journal publisher.

Abstract

In-memory neural network inference is limited by the energy and area overhead of analog-to-digital converters. Spiking neural networks (SNNs) implemented using in-memory computing mitigate this by producing spike-based outputs, but they are commonly trained using surrogate gradient descent, which is computationally expensive, sensitive to hyperparameters, and difficult to enforce hardware constraints or to adapt to changes in neuron dynamics during training. This work proposes a low-complexity, simulation-aware optimization method that tunes heterogeneous absolute refractory periods in SNNs initialized from trained ANN weights. A rule-based greedy strategy is used to regulate neuron spiking without modifying synaptic weights. Circuit simulations show negligible power overhead from refractory tuning, while experimental results demonstrate up to 7 5 - 8 0 % spike reduction compared to a baseline SNN with a unity refractory period, achieving accuracy comparable to or exceeding that of the parent ANN.

Citation Information

Text

author      S. N. B. Tushar and S. H. Alam and N. N Chakraborty and C. D. Schuman 
            and H. Das and  G. S. Rose
title       Low-Complexity Rule-Based Greedy Optimization of Refractory Periods for In-Memory SNN
booktitle   IEEE 19th Dallas Circuits and Systems Conference (DCAS)
url         https://ieeexplore.ieee.org/abstract/document/11544440
month       April
year        2026
doi         10.1109/DCAS69364.2026.11544440

Bibtex

@INPROCEEDINGS{tac:26:lcr,
    author = "S. N. B. Tushar and S. H. Alam and N. N Chakraborty and C. D. Schuman 
                and H. Das and  G. S. Rose",
    title = "Low-Complexity Rule-Based Greedy Optimization of Refractory Periods for In-Memory SNN",
    booktitle = "IEEE 19th Dallas Circuits and Systems Conference (DCAS)",
    url = "https://ieeexplore.ieee.org/abstract/document/11544440",
    month = "April",
    year = "2026",
    doi = "10.1109/DCAS69364.2026.11544440"
}