AI Hardware for Edge Intelligence

In the UMass Amherst Nanodevices and Integrated Systems Lab, professor of electrical and computer engineering Qiangfei Xia leads a team making breakthrough devices that could lead to smaller, smarter, faster computer chips and reshape the fields of artificial intelligence (AI) and machine learning (ML).

A memristor is an electrical component that controls the flow of electrical current in a circuit, while also “remembering” the prior state, even with the power turned off. In this way, memristors differ from transistors—currently the main component of computer chips—which lose all information once the current is interrupted.

Xia and his colleagues have notched several groundbreaking achievements: integrating the largest analog memristive crossbar arrays (at the time of publication) and demonstrating a broad spectrum of machine intelligence applications; fabricating the world’s smallest memristive devices (two nanometers) in a crossbar circuit; making the tallest memristor 3D crossbar array; and most recently devising a process for removing “noise” from memristors, thus producing record-high levels of conductance—and underscoring the feasibility of memristor arrays for large-scale commercial applications.

Under an National Science Foundation grant, Xia has been developing a new hardware system that integrates two different types of memristors and supporting circuits into 3D networks. In a way, these 3D memristor networks mimic the behavior of the brain—they can both compute and store data at the same site—making them a potentially ideal fit for real-time AI and ML computing.

Xia’s lab is collaborating with the Defense Advanced Research Projects Agency (DARPA) to develop a camera that tracks events more quickly by only transmitting data from pixels that have detected changes. Xia’s work is part of DARPA’s effort to create a “neuromorphic camera” in which each pixel operates independently. This would pave the way for more intelligent sensors that can be applied to robotics, autonomous vehicles, and beyond.

Xia also co-leads the AI Hardware topical area for the Northeast Microelectronic Coalition (NEMC), one of eight regional innovation hubs in the nation under the Microelectronics Commons program.

“Our goal is to bring intelligence to the edge and process data at the site where they are generated with new hardware that has advantages in size, weight, and power consumption,” Xia says. “Such an approach will circumvent constant data shuttling, reduce latency, and mitigate security risk for AI and ML applications.”