ReckOn - Charlotte Frenkel

Learn about Charlotte Frenkel's neuromorphic hardware: ReckOn

ReckOn At A Glance

Release Year: 2022
Status: Released
Chip Type: Digital
Applications: Task-agnostic learning over second-long timescales at the extreme edge
Neurons: 256
Synapses: 256
Weight bits: 8
Activation bits: 16
On-Chip Learning: true
Power: 20µW - 150µW (from real-time to accelerated-time learning) at 0.5V

ReckOn demonstrates, for the first time, end-to-end on-chip learning over second-long timescales (no external memory accesses, no pre-training). It is based on a bio-inspired alternative to backpropagation through time (BPTT), the e-prop training algorithm, which has been modified to reduce the memory overhead required for training to only 0.8% of the equivalent inference-only design. This allows for a low-cost solution with a 0.45-mm² core area and a <50-µW power budget at 0.5V for real-time learning in 28-nm FDSOI CMOS, which is suitable for an always-on deployment at the extreme edge. Furthermore, similarly to the brain, ReckOn exploits the sensor-agnostic property of spike-based information. Combined with code-agnostic e-prop-based training, this leads to a task-agnostic learning chip that is demonstrated on vision, audition and navigation tasks.

Overview

ReckOn, a spiking recurrent neural network (RNN) processor, enables on-chip learning over second-long timescales using a modified e-prop algorithm. Developed by Charlotte Frenkel at the Institute of Neuroinformatics, and fabricated in 28-nm FDSOI CMOS, it was presented at the 2022 IEEE International Solid-State Circuits Conference (ISSCC), marking it as the first spiking neuromorphic chip at this forum. Distinctively, ReckOn demonstrates on-device learning for extended durations on real-world tasks without external memory, a notable advancement in neuromorphic computing. The processor is fully open-source and has been applied in gesture recognition, keyword spotting, and navigation, adaptable to various spike-encoded sensory modalities. While the actual chips are not commercially available, ReckOn can be implemented on small-scale FPGAs. The design, primarily self-contained with input and output layers, does not support multi-core large-scale integrations but can be adapted for such purposes. Documentation and Verilog source files are accessible at https://github.com/ChFrenkel/ReckOn/ .

DateTitleAuthorsVenue/Source
March 2022ReckOn: A 28nm sub-mm² task-agnostic spiking recurrent neural network processor enabling on-chip learning over second-long timescalesC. Frenkel and G. IndiveriIEEE International Solid-State Circuits Conference (ISSCC)

Help Us Improve this Guide

Our hardware guide is community-maintained. If you know of a chip we should add, see an error, or have updated information, please let us know by opening an issue on our GitHub repository.

Suggest an Edit on GitHub
NorthPole, IBM's latest Neuromorphic AI Hardware

NorthPole, IBM's latest Neuromorphic AI Hardware

A deep dive into IBM's NorthPole, a brain-inspired AI accelerator. Understand its architecture, 10 core axioms, and how it achieves groundbreaking energy efficiency for neural inference.

Digital Neuromorphic Hardware Read List

Digital Neuromorphic Hardware Read List

Stay up-to-date with cutting-edge digital hardware designs for neuromorphic applications. Explore recent research on power-efficient event-driven spiking neural networks and state-of-the-art processors like TrueNorth and Loihi.

Spiking Neurons: A Digital Hardware Implementation

Spiking Neurons: A Digital Hardware Implementation

Learn how to model Leaky Integrate and Fire (LIF) neurons in digital hardware. Understand spike communication, synapse integration, and more for hardware implementation.