On the Link Between Energy & Information for the Design of Neuromorphic Systems
summary
Argument links energy and information to build edge neuromorphic hardware: Eta Compute demonstrates an asynchronous, event-driven chip in 55 nm that supports ~2k neurons and ~128k synapses, runs at up to 60 MHz while drawing ~2 mW and can scale to microW at 0.2 V operation. On-chip stack mixes spiking SNNs, a Cortex-M3 and DSP, uses 8-bit synapses and plasticity, and achieves ~96% on 10-digit and ~85% on 30-word speech tasks using ~500 neurons and ~25k synapses, showing the approach works.
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:16Edge Devices and their relationship with the Cloud
- 2:50Eta Compute is focused on addressing issues at the edge using neuromorphic computing
- 5:52Brain evolved to balance energy consumption with information processing • Brain is unique-serves as an interface between
- 6:55Bulk of energy consumed for spike generation & for its speed, quality and propagation needs
- 7:41Asynchronous Spike Timing Maximizes Information Rates
- 12:43Biophysics at synapse optimized for energy efficient information processing
- 15:09Synaptic plasticity enables optimal filtering of information for efficient information transfer
- 15:59Population Dynamics Biased to Optimize energy while enhancing information processing
- 16:46Massively Parallel & Asynchronous Brain Enables System Level Efficiency during information processing
- 17:30Strong link between energy & information offers a blueprint for the design of neuromorphic systems
- 19:16Eta Compute Asynchronous Single Core
- 23:50Eta Compute Asynchronous Multicore Chip
- 25:53Example: Continuous Speech Recognition • Trained using Google Command Word Set 2300 speakers and
- 29:51Example: MNIST Character Recognition
- 32:15Summary
related talks
01h 57mIntroduction to Neuromorphic Computing
41mBrain-inspired computational primitives for reliable neuromorphic intelligence
43mKeeping Moore's law alive: Neuromorphic computing
20mRecap and Closing Remarks: Neuropunk revolution and energy consumption
01h 16mRethinking Computing with Neuro-Inspired Learning: Algorithms to Hardware
20mLow Power AI Edge Hardware
01h 14mTissue vs silicone musings on the future of deep learning hardware and software
49mNew Computer Architectures : Explore Quantum Computers & SyNAPSE neuromorphic chips