EE Times Current

EE Times Current comes from EE Times, and in practice it carries two different kinds of program under one name. The first is Brains and Machines, a series on neuromorphic engineering and biologically inspired technology. Taylor Marvin opens these episodes, and the conversations belong to Dr. Sonny Baines of University College London and Dr. Giulia D'Angelo of the Czech Technical University in Prague. Baines does the interviewing. Guests move freely between the lab, the startup and the newsroom. Professor Rodolphe Sepulchre of the University of Cambridge brings control theory into the discussion. He describes hybrid systems, which package differential equations and discrete logic together, and then flags their main limitation. Professor Chris Eliasmith's work took him from basic research and cognitive modeling to helping spin out a low-power AI chip. He explains why time-sensitive problems favor timing codes while noisy ones tend to favor rate codes. He also argues for sizing a population of neurons to the precision a task requires, which sometimes means a population of two. Neuronova's two co-founders, its CEO and CTO, join, and one of them offers a rule of thumb for low-power hardware. Judge a chip by what it consumes when it is off. Sally Ward-Foxton, who covers AI for EE Times, brings the AI chip market into the series. That conversation reaches the question of what kind of learning a robot can support without a big Nvidia GPU, then moves to latency in large data centers, and the same episode follows the interview with a guest from Johns Hopkins University on the issues it raised. The second strand is more applied, and it is sponsored. Eric Singer introduces these episodes, made with support from Siemens EDA or Synopsys, and the guests come from those companies. The Siemens EDA pair describe push-button thermal analysis, meant to spare designers from becoming thermal experts. They also say timing closure on a 3D IC is still achievable but takes far more work, since power, heat and stress all have to be understood at once. Hezi Tsar of Synopsys calls the car a physical AI device at the edge and a test case for multi-die designs, where partitioning workloads into dies and chiplets is only one of the problems. Across both strands the talk runs on working terms: rate codes, hybrid systems, timing closure, ADAS, chiplets. What a listener gets is the reasoning of the people building this hardware, in their own words rather than condensed into a news item. 'Can the Nvidia Monopoly on AI Chips Be Broken?' is a sensible first listen, since it sets the commercial chip market against the neuromorphic series' questions about learning and latency. Episodes point to eetimes.com for transcripts and links to the stories mentioned.