Compute-in-memory implementation of state space models for event sequence processing
Xiaoyu Zhang1, Mingtao Hu1, Sen Lu1
1Department of Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, MI, USA.
Nature Communications
|January 9, 2026
Summary
State space models (SSMs) are now efficient on new hardware. This research integrates SSMs with compute-in-memory hardware for real-time, event-driven processing in AI tasks.
Area of Science:
- Artificial Intelligence
- Computer Engineering
- Neuroscience
Background:
- State space models (SSMs) offer advanced long sequence processing capabilities.
- SSMs generalize recurrent and convolutional networks, mimicking biological system functions.
- Existing SSM implementations face challenges in energy efficiency and real-time processing.
Purpose of the Study:
- To implement state space models (SSMs) on energy-efficient compute-in-memory hardware.
- To achieve real-time, event-driven processing for AI applications.
- To explore algorithm and hardware co-design for enhanced performance.
Main Methods:
- Re-parameterized SSMs for real-valued coefficients and shared decay constants.
- Leveraged device dynamics and diagonalized state transition parameters.
- Implemented state evolution natively in crossbar-based compute-in-memory systems with memristors.
Main Results:
- Achieved high accuracy in AI tasks using the proposed system.
- Demonstrated significant energy efficiency compared to traditional methods.
- Enabled fully asynchronous processing for event-based vision and audio.
Conclusions:
- Algorithm and hardware co-design enables efficient SSM implementation.
- The system provides a pathway for real-time, low-power AI processing.
- This approach is suitable for event-based sensory data tasks.
More Related Videos
09:46MPI CyberMotion Simulator: Implementation of a Novel Motion Simulator to Investigate Multisensory Path Integration in Three Dimensions
Published on: May 10, 2012
13.1K
06:09P300-Based Brain-Computer Interface Speller Performance Estimation with Classifier-Based Latency Estimation
Published on: September 8, 2023
922
Related Concept Videos
01:27State Space Representation
519
The frequency-domain technique, commonly used in analyzing and designing feedback control systems, is effective for linear, time-invariant systems. However, it falls short when dealing with nonlinear, time-varying, and multiple-input multiple-output systems. The time-domain or state-space approach addresses these limitations by utilizing state variables to construct simultaneous, first-order differential equations, known as state equations, for an nth-order system.
Consider an RLC circuit, a...
Consider an RLC circuit, a...
519
01:29Mechanistic Models: Compartment Models in Algorithms for Numerical Problem Solving
282
Mechanistic models play a crucial role in algorithms for numerical problem-solving, particularly in nonlinear mixed effects modeling (NMEM). These models aim to minimize specific objective functions by evaluating various parameter estimates, leading to the development of systematic algorithms. In some cases, linearization techniques approximate the model using linear equations.
In individual population analyses, different algorithms are employed, such as Cauchy's method, which uses a...
In individual population analyses, different algorithms are employed, such as Cauchy's method, which uses a...
282
01:21State Space to Transfer Function
552
The conversion of state-space representation to a transfer function is a fundamental process in system analysis. It provides a method for transitioning from a time-domain description to a frequency-domain representation, which is crucial for simplifying the analysis and design of control systems.
The transformation process begins with the state-space representation, characterized by the state equation and the output equation. These equations are typically represented as:
The transformation process begins with the state-space representation, characterized by the state equation and the output equation. These equations are typically represented as:
552
01:26Per-Unit Sequence Models
426
An ideal Y-Y transformer, grounded through neutral impedances, displays per-unit sequence networks akin to those of a single-phase ideal transformer when subjected to balanced positive- or negative-sequence currents. These currents do not produce neutral currents, and their associated voltage drops.
Zero-sequence currents, which are identical in magnitude and phase, generate a neutral current, resulting in voltage drops across the neutral impedance and the low-voltage winding. If the...
Zero-sequence currents, which are identical in magnitude and phase, generate a neutral current, resulting in voltage drops across the neutral impedance and the low-voltage winding. If the...
426
01:23Transfer Function to State Space
748
State-space representation is a powerful tool for simulating physical systems on digital computers, necessitating the conversion of the transfer function into state-space form. Consider an nth-order linear differential equation with constant coefficients, like those encountered in an RLC circuit. The state variables are selected as the output and its n−1 derivatives. Differentiating these variables and substituting them back into the original equation produces the state equations.
In an RLC...
In an RLC...
748
01:23Linear time-invariant Systems
863
A system is linear if it displays the characteristics of homogeneity and additivity, together termed the superposition property. This principle is fundamental in all linear systems. Linear time-invariant (LTI) systems include systems with linear elements and constant parameters.
The input-output behavior of an LTI system can be fully defined by its response to an impulsive excitation at its input. Once this impulse response is known, the system's reaction to any other input can be...
The input-output behavior of an LTI system can be fully defined by its response to an impulsive excitation at its input. Once this impulse response is known, the system's reaction to any other input can be...
863