Sandia Labs FY21 LDRD Annual Report

FY21 ANNUAL REPORT

LDRD IMPACT STORY: Hardware Acceleration of Adaptive Neural Algorithms (HAANA) Grand Challenge (2015-2017)

Situations are steeped in data. When computers are presented with a certain set of circumstances, microprocessors utilize programmed algorithms to synthesize the input and identify the best action to take in response. But as more variables are introduced, the data set grows exponentially, slowing down the system and response time. To solve the mounting problem created by big data, Sandia researchers, led by PI Conrad James, and academic partner, the University of Illinois Urbana-Champaign, conceptualized and developed a radical new approach. The new system integrated neural-inspired machine-learning algorithms with neural-inspired architecture cores, in conventional technology, allowing it to comb through large volumes of data and identify specific data patterns focused precisely on the activity of interest. The initial success of the endeavor, a Temporal Processing Unit, provided the nation with a mechanism for rapidly transforming raw cyber data into information for real-time, adaptive recognition and response, but the new technology also provided a neural network categorizer that spun off numerous other deliverables. Since 2015, the HAANA Grand Challenge has resulted in 19 worldwide patents, with 10 occurring in 2021 – one of those belonging to Sandia researchers, Matt Marinella and Sapan Agarwal (who originally hired in as a postdoctoral researcher to work on the HAANA Grand

Challenge). Agarwal said of his experience, “When I joined Sandia, I was a primarily an expert in semiconductor devices. Through the HAANA GC, I was able to learn from neural algorithm and architecture experts and became an expert in multi-scale codesign of computing systems from material to algorithms. This [project] has provided me the opportunity to subsequently lead projects in the co-design of heterogenous radiation hardened systems, analog neural accelerators, solvers for linear systems and more.” The HAANA Grand Challenge has also resulted in more academic collaborations including a three-year LDRD project with the University of Texas at Austin, other workshops, and 39 publications directly associated with the effort.

“

Through the HAANA GC, I was able to learn from neural algorithm and architecture experts and became an expert in multi-scale codesign of computing systems from material to algorithms.

“

76

LABORATORY DIRECTED RESEARCH & DEVELOPMENT