EE Times Current · 5 Dec 2025 · 55m
Professor Chris Eliasmith of the University of Waterloo discusses his Semantic Pointer Architecture Unified Network (SPAUN) and the Neural Engineering Framework (NEF), explaining how they enable cognitive modeling and efficient neuromorphic computing. He details the development of Legendre Memory Units (LMUs) and the spin-out company Applied Brain Research, which recently released a low-power MRAM-based chip for state-space models. The episode concludes with a debate between hosts Dr. Sunny Bains and Dr. Giulia D’Angelo, and commentator Prof. Ralph Etienne-Cummings, regarding the efficiency, standardization, and practical utility of these neuromorphic approaches.
Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
Engineering & simulationLegendre Memory Units (LMU)650 times fewer parameters to get the same performance as an LSTM
The team developed Legendre Memory Units to represent time series data, which outperformed LSTMs and Transformers by using significantly fewer parameters for the same performance.
“we found tasks where the LMU would use 650 times fewer parameters to get the same performance as in LSTM. So just massive difference.”
Speaker 4 · Hear it at 25:54
ManufacturingApplied Brain Research (ABR) chipunder 30 millowatts for full speech recognition
The team fabricated a custom chip using MRAM to natively run state-based models, achieving world records in low-power operation for speech recognition.
“we just got the chip back for it at the end of August. We got the chip back. We tested it and we had it up and running and setting world records in sort of low power running state-based models within a week and a half…”
Speaker 4 · Hear it at 32:21
early-stage companies and companies scaling up
Other shows on the same ground, closest first.
Hosts Chris Gammell and Dave Jones mourn the recent deaths of Neil Armstrong and Hans Camenzind, reflecting on their engineering legacies and the state of space exploration funding. They discuss the Spinnaker high-performance computing chip designed to model the human brain, noting its 57,600 ARM cores and the economic barriers to further miniaturization. The conversation shifts to practical hardware development, contrasting breadboarding with direct PCB prototyping for modern surface-mount components, and highlighting Osram's initiative to standardize LED footprints for easier second-sourcing.
Hosts Daniel and Forbode discuss NeuralTree, a neurochip developed by researchers at EPFL (École Polytechnique Fédérale de Lausanne) designed to detect and counter neural disorders like epilepsy and Parkinson's disease. The chip features 256 input channels, an area of 3.5 square millimeters, and an energy-aware machine learning algorithm that allows it to operate efficiently on the edge without cloud dependency. The hosts highlight the device's ability to distinguish between healthy and irregular neural impulses, its potential for self-evolving algorithms to adapt to changing brain signals, and its broader implications for neurological research and treatment.
Ben Bajarin and Jay Goldberg host Austin Lyons and Paul Karazuba to discuss the architecture, market, and future of Neural Processing Units (NPUs). The group debunks the metric 'TOPS per watt' as meaningless, arguing that real-world performance depends on specific network workloads and latency. They explore how chipmakers like Qualcomm, Intel, and Apple are designing NPUs, often by licensing IP or evolving DSPs, and predict that NPUs will become ubiquitous in devices from smartphones to microcontrollers within three to four years.
Nandan Nayampally, CMO at BrainChip, discusses the evolution of AIoT and the shift of AI processing from the cloud to the edge. He explains how neuromorphic computing enables real-time, private, and efficient inference on billions of devices, citing PwC's $15 trillion GDP impact estimate. Nayampally highlights key adoption markets like healthcare, agriculture, and infrastructure, and advises companies to focus on solving acute pain points ('headaches') rather than nice-to-have features ('vitamins') to gain traction.
Hackaday editors Mike Szczys and Elliot Williams discuss the technical realities and hype surrounding Neuralink's brain-computer interface, analyzing the gap between their surgical robot demo and actual neuroscience. They also cover the 40th anniversary of the Sony Walkman, arguing that its success was driven by the design of its headphones rather than the player itself. The episode highlights several hardware hacks, including synthesizing 7400-series logic from Verilog, a subtractive 3D printing process using MDF, and a rotary phone hack that controls web browsers.
The summary, chapters and topics are generated from the episode’s audio by this directory, not written by EE Times Current. The episode itself belongs to its publisher.