“We cannot just build on the platforms we already have. We must build new and different technologies that are considered disruptive or ‘unconventional’.”
- Joshua Yang
“In this era of big data and the internet of things, we are faced with tons of data. Our devices must be able to process things faster and yet with a lower energy,” says Yang. He notes, however, that current computing technology is not up to the task of processing the growing mounds of data efficiently and with sustainability in mind.
“We cannot just build on the platforms we already have. We must think differently. We must build new and different technologies that are considered disruptive or ‘unconventional’,” says Yang.
Yang believes neuromorphic computing, configuring microprocessors to mimic aspects of the human brain, holds much promise for taking computing beyond its current energy efficiency and processing limitations.
“We are looking at how human brains do information processing and storage. We want to build something with real intelligence, computers that can really think and learn, not just use software and human-programmed algorithms,” says Yang. To meet that challenge, he and his colleagues are working on hardware designs that can learn under the same principles used by synapses and neurons in the brain.
One of Yang’s recent breakthroughs, published in the journal Nature Materials is a first step toward meeting that challenge. He and his colleagues have developed a very tiny electrical resistance switch called a diffusive memristor that can emulate synapses faithfully, the place where signals pass through from one nerve cell to another in the human brain. Memristors are devices that can store and process information while offering several key performance characteristics that exceed conventional integrated circuitry. They don’t require exotic materials or high temperatures in their manufacture and they can be used for different purposes such as memory storage, information processing, security, and as a sensor.
“Memristive technology is simple yet versatile. Memristors can be made at a very small scale—10,000 times smaller than the width of a human hair. They can also be stacked, something you can’t do with most silicon devices,” says Yang.
Universal memory is one application Yang would like to see memristive technology used for as well. “Right now we have memory hierarchies, different types of memory with different attributes and performance. Instead, we need a universal memory designed to be good at everything – to be fast, dense, have low energy requirements, and be non-volatile. Universal memory means a much simpler computer in terms of components, it will use less energy and store more information and will be faster. We will be able to process the same information with orders of magnitude less energy and faster speed,” says Yang.
Though Yang and his colleagues have had some exciting breakthroughs, he says their neuromorphic research is at a very early stage. Their recent paper on synaptic emulators for diffusive memristors is just a small part of the story.
“It’s just one building block. We now want to emulate a neuron, then integrate synapses and neurons together to build a neural network, that’s what’s next. We will pick the neuroscientists’ brains to get their latest knowledge to implement in our electronic platform. This will help the neuroscientists, too. We may have a better platform to verify their theories of the brain than animal tissues, which is still pretty much a black box. Our electronic system is well defined and can be checked. It can help us to get answers. In addition, it is also a great natural platform for novel computing paradigms,” says Yang.
Before coming to campus in 2015, Yang spent eight years at Hewlett-Packard Labs exploring new computing devices and approaches. He holds 77 granted and over 70 pending patents, most of which have been licensed by and technology-transferred to industry for product development. He has authored and co-authored over 100 well-cited papers in peer-reviewed academic journals for computer engineering technologies developed throughout his career so far. With his applied science mindset and his materials science background (MS and PhD from University of Wisconsin-Madison Materials Science Program), Yang knew he needed interdisciplinary expertise to pursue transformational breakthroughs in neuromorphic computing.
“What brought me to UMass was the strengths here in materials, devices, electrical engineering and very top artificial intelligence and polymer science programs. For instance, my colleague Professor Qiangfei Xia is a world-leading expert in nanodevice fabrication and integration. UMass also has strong activity in bio research, which has recently been further strengthened by the new Institute for Applied Life Sciences (IALS),” Yang notes.
Yang and his interdisciplinary team of colleagues are definitely on to something. In only two years, Yang has secured $3.2 million dollars of external funding as principal investigator to move this research forward. “In engineering schools we need to ask ourselves if what we are investigating is useful or not. Application-based research helps to get funding. Plus, we have a very strong team and great support from the department and the school as well,” smiles Yang.
Karen J. Hayes '85