Runtime Tunable Tsetlin Machines for Edge Inference on eFPGAs

TL;DR

This paper introduces a runtime-tunable Tsetlin Machine accelerator for eFPGAs that minimizes resource usage, allows on-field model adjustments, and significantly reduces energy consumption compared to microcontrollers.

Contribution

It presents a novel eFPGA accelerator for Tsetlin Machines enabling runtime model tuning and resource-efficient inference without offline resynthesis.

Findings

Uses 2.5x fewer LUTs and 3.38x fewer registers than previous designs.

Achieves up to 129x energy reduction compared to microcontrollers.

Supports flexible model size, architecture, and input data changes at runtime.

Abstract

Embedded Field-Programmable Gate Arrays (eFPGAs) allow for the design of hardware accelerators of edge Machine Learning (ML) applications at a lower power budget compared with traditional FPGA platforms. However, the limited eFPGA logic and memory significantly constrain compute capabilities and model size. As such, ML application deployment on eFPGAs is in direct contrast with the most recent FPGA approaches developing architecture-specific implementations and maximizing throughput over resource frugality. This paper focuses on the opposite side of this trade-off: the proposed eFPGA accelerator focuses on minimizing resource usage and allowing flexibility for on-field recalibration over throughput. This allows for runtime changes in model size, architecture, and input data dimensionality without offline resynthesis. This is made possible through the use of a bitwise compressed inference architecture of the Tsetlin Machine (TM) algorithm. TM compute does not require any multiplication operations, being limited to only bitwise AND, OR, NOT, summations and additions. Additionally, TM model compression allows the entire model to fit within the on-chip block RAM of the eFPGA. The paper uses this accelerator to propose a strategy for runtime model tuning in the field. The proposed approach uses 2.5x fewer Look-up-Tables (LUTs) and 3.38x fewer registers than the current most resource-fugal design and achieves up to 129x energy reduction compared with low-power microcontrollers running the same ML application.