//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>
On April 26, California Institute of Technology Emeritus professor Carver Mead—best known for starting the very-large-scale integration (VLSI) chip revolution with Lynn Conway—was presented with a special lifetime achievement award for his work in neuromorphic engineering: building brain-like circuits in silicon. Already the winner of dozens of honors and awards including the National Medal of Technology in 2002, he accepted the Mahowald Prize at a ceremony as part of the Neuro Inspired Computational Elements Conference (NICE 2024) at the University of California at San Diego (UCSD).
“Carver Mead established the field of Neuromorphic Electronic Engineering. His creativity and vision has inspired a generation of scientists, technologists and entrepreneurs to emulate brain-like information processing in electronic systems,” according to the Mahowald Prize jury, led by Terry Sejnowski of UCSD
The main (annual) prize was presented to the Neural Exploration & Research Laboratory from Sandia National Laboratories lead by Brad Aimone. “The Sandia team demonstrated that neuromorphic hardware can efficiently implement Monte Carlo methods for solving differential equations,” the jury said. These are used “for solving a wide range of problems including those in heat transfer, medical imaging and finance.”
Mahowald and Mead
Misha Mahowald, for whom the field was named, was herself a pioneer in the field, with her Ph.D. thesis winning the 1992 Clauser prize set up to reward originality, and specifically, “potential for opening up new avenues of human thought and endeavor as well as by the ingenuity with which it has been carried out.” After she left Caltech, she and several other colleagues started a neuromorphic group at the Institute of Neuroinformatics, part of the ETH Zürich and University of Zürich.
By MRPeasy 05.01.2024
By Global Unichip Corp. 04.18.2024
In his acceptance speech, Mead explained that, although he is usually credited with founding the field, he and Mahowald really did it together. Speaking in an interview for the Brains and Machine podcast, to air later this year, Mead said that he spent the 1970s being frustrated that no-one was working on his main interest: parallel computation in the VLSI he had spent his life helping to study and progress.
In 1980, he was venting at lunch with two friends—Richard Feynman and John Hopfield—that there must be paradigms for parallel computation. At that point, he asked himself, and them, what happens in the brains of animals? This led to the development of the Computation and Neural Systems course, which the three taught jointly and Mahowald ended up taking.
“[Mahowald] was a biology major and had been working with a group that had gotten into the retina and figured it out. She was just reading all the literature on retinas and how they were wired and trying to imagine what was going on in that wiring,” Mead said. “She joined our group and was teaching me what she was learning about the retina…and so I got jazzed about it.
“If you have a theory, you don’t really know if it’s right unless you can make it work somehow. And these things were complicated enough that simulating them on the computers of the day were hopelessly slow,” he added. “Doing it on a circuit where you could make a real physical object that did some version of what you were thinking the retina did. So, it was really this freshman who had come to the course…and had the sense to say, ‘Here’s a thing that we know a lot about the inputs and outputs, but we don’t know how it does it. We could become clear…by making something that does that.’
“[Mahowald] was a very deep thinker,” Mead concluded. Unfortunately, she died in tragic circumstances in 1996.
A mathematical approach
Brad Aimone, leader of the prizewinning Sandia team, recognizes that his group’s work is out of the mainstream of neuromorphic engineering, but sees it as important. “The reason that the Department of Energy does this is for a wide range of computational physics applications. That’s why we build supercomputers. And what’s happened in the last decade in the supercomputer world is there’s an increased emphasis on GPUs, just like everywhere else, [and] data centers in general.
“GPUs are really good at some things and not very good at others,” he said. “And one of the things they’re not particularly good at is random Monte Carlo sampling. That’s why we targeted that application.”
Asked why he wanted to do this kind of work given that his own Ph.D. was actually in neuroscience, Aimone said, “I kind of see that there’s a parallel looking for large scale computational physics simulations and the needs of the Department of Energy. The circuit is not biological, the task itself is not bio inspired, but the fact that the brains do complicated arithmetic seems obvious to me. The fact that brains can do things at large scales better than what we can seems like a very important hypothesis to go after.”
The Misha Mahowald prizes for neuromorphic engineering have been awarded every year since 2016, but only once before has the jury recognized a lifetime contribution to the field: for Karlheinz Meier of Heidelberg University in Germany, leader of the group that created BrainScaleS.
