Compute-in-Memory Implementation of State Space Models for Event Sequence Processing

2025-11-17

AI Summary

State-space models (SSMs) suffer from low energy efficiency and high synchronization overhead when deployed on compute-in-memory (CIM) hardware for real-time, event-driven long-sequence processing. Method: This paper proposes an algorithm–hardware co-design approach: (i) mapping SSM state evolution onto the short-term memory dynamics of memristive crossbars; (ii) applying real-coefficient reparameterization with shared diagonal decay constants to enable hardware-friendly parameter compression and diagonalization; and (iii) enabling fully asynchronous, event-triggered state updates—the first such support in CIM architectures. Contribution/Results: The design intrinsically unifies computation and storage, eliminating the von Neumann bottleneck. Evaluated on event-camera and spike-audio tasks, it achieves state-of-the-art accuracy while improving energy efficiency by 2.3–4.1× and reducing latency by 68%, establishing a new paradigm for low-power edge sequence intelligence.

Technology Category

• Machine Learning: Hardware-aware ML
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Learning & Optimization for CV

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search

Abstract

State space models (SSMs) have recently emerged as a powerful framework for long sequence processing, outperforming traditional methods on diverse benchmarks. Fundamentally, SSMs can generalize both recurrent and convolutional networks and have been shown to even capture key functions of biological systems. Here we report an approach to implement SSMs in energy-efficient compute-in-memory (CIM) hardware to achieve real-time, event-driven processing. Our work re-parameterizes the model to function with real-valued coefficients and shared decay constants, reducing the complexity of model mapping onto practical hardware systems. By leveraging device dynamics and diagonalized state transition parameters, the state evolution can be natively implemented in crossbar-based CIM systems combined with memristors exhibiting short-term memory effects. Through this algorithm and hardware co-design, we show the proposed system offers both high accuracy and high energy efficiency while supporting fully asynchronous processing for event-based vision and audio tasks.

Problem

• Implementing state space models in energy-efficient compute-in-memory hardware
• Reducing model complexity for practical hardware mapping
• Enabling real-time asynchronous processing for event-based tasks

Innovation

• Compute-in-Memory hardware for State Space Models
• Reparameterized model with real-valued coefficients
• Crossbar-based systems with memristors for asynchronous processing