Beyond Von Neumann: Applied Materials and Arm Lead DARPA-Funded Research on a Neuromorphic Switch for AI
Back in July, Applied Materials announced that we’d been selected by the Defense Advanced Research Projects Agency (DARPA) to develop technology for AI. While Applied is engaged on the development of many disruptive technologies, it’s not often that we’re in a position to discuss them in early development. Thanks to the vision of DARPA’s Electronics Resurgence Initiative and their generous investments to take us beyond Von Neumann computing, we’re in a position to pursue some real game-changing device technology. We can also share a few more details about what early technology development looks like at Applied.
At the core of this project are the efforts of Applied, Arm and Symetrix to develop a correlated electron switch. This of course begs the question, “what the heck is a correlated electron?” Correlated electron materials are materials that break some of the rules of classical band theory. These materials are really best considered as a type of quantum matter. Most of the materials we are studying in the context of this research should, according to classical theories, conduct electrons, but thanks to correlation they can also exhibit insulating properties under certain conditions due to electron-electron interactions that don’t happen according to traditional band theory. Welcome to the weird and wonderful world of quantum matter!
Many of the other technologies looking at conductor-insulator switching necessarily require the motion of atoms and, in many cases, accompanying phase transitions. The promise of integrating correlated materials is that we’re looking at a purely electronic phase transition. We’re optimistic that if we can pin down the right material to integrate, then this will lead to unprecedented switching speeds since no atom migration is required. It’s also likely that reliability and endurance (how many times you can switch the device on and off before it fails) will be improved since a primary mode of failure involves atoms getting “stuck” at interfaces and defects when atomic motion is the basis of switching. Finally, the ability of the device to function at sub 1K temperatures is also embraced in the theory of these materials since we don’t need background kinetic energy to drive atomic motion.
I was led to this branch of materials and device research thanks to an introduction to Carlos Paz de Araujo, chairman and founder of Symetrix by another friend, Lucian Shifren of Arm Research. Carlos and his team have been evaluating several materials that are showing the promise of integrating correlated materials that demonstrate the concept and feasibility of these devices. He’s also been busy at work developing the underlying theory behind the operation of the device.
At the heart of these new devices is the integrated quantum matter — in this case the correlated electron material. Manipulation of the composition of the material (the percentage and types of atoms), the phase and morphology (e.g. roughness and grain size of crystalline regions) and establishing the right interfaces at the electrodes connected to the material are at the heart of what Applied Materials does best. Suffice it to say, it didn’t take long before the three parties saw real value in getting together on developing this new technology. Symetrix brings the general concept of the correlated electron switch and early learnings; Arm brings circuit designs and clearly defined requirements on how we need to optimize the device characteristics to disrupt compute at the edge with a disruptively low-power chip; and Applied does what we do best by demonstrating the path to manufacturing through enabling new materials and integration for a new device.
We also augmented our team with Dan Dessau from the University of Colorado Boulder and George Sawatzky from the University of British Columbia. They each bring specialized surface science and modeling techniques to our collaboration team that should allow us to effectively probe the quantum structure of the materials as we switch them.
It’s a privilege to have had our team selected by DARPA to pursue this technology and now the real work begins. These journeys are rarely linear and all new technology development has unexpected twists and turns. I’ll update you again in the relatively near future on why we think this technology is a disruptor for AI, and will keep the community up to date on our efforts as our team endeavors to make possible the technology shaping the future.
David Thompson is the Primary Investigator and Applied Materials’ Team Leader for the “Synapse and Neuron Correlated Electron Devices” Research Project funded in part by the Defense Advanced Research Projects Agency (DARPA). His bio can be found here. He also leads our Center of Excellence in Chemistry and works with our product development teams to incorporate new processes into our broad portfolio of products. Over the next several months we will feature updates from David on the technology development of our DARPA program.