Paper Title: Enabling sophisticated Endpoint AI by leveraging SPOT and MRAM

Paper Abstract: AI is a memory-intensive application. Its storage requirements range from hundreds of kilobytes to hundreds of gigabytes, and memory access is the largest consumer of power during an AI computation. This presentation will introduce the impact of these constraints on Endpoint AI applications, and the suitability and practicality of using a combination of MRAM, tightly-coupled memory, and dedicated SRAM based on a series of experiments measuring power and performance of various memory configurations.

Paper Author: Carlos Morales, VP of AI, AMBIQ

Author Bio: Carlos Morales has over 30 years of research and development experience spanning silicon to cloud. During his career, Mr. Morales has held architecture, development, management, and business development roles at multiple technology companies, ranging from large multi-nationals to pioneering silicon startups. He is currently the Vice President of Artificial Intelligence at Ambiq, where his team’s mission is to leverage SPOT to bring Endpoint AI to the farthest reaches of edge compute. Carlos joins us from Intel, where he ran the AI Product Group’s global AI software organization, and led Intel-wide AI strategy and business development initiatives. Before joining Nervana Systems (acquired by Intel), Carlos was part of Cisco’s Advanced Architecture and Research group, where he pioneered novel technologies such as Fog Computing, and the first full-featured software-defined router. Carlos’ career spans just about every aspect of computer engineering, from ASIC development to cloud-based Deep Learning-as-a-Service products. Besides AI, his past projects included Cloud-based back-end applications, cybersecurity, workload scheduling, orchestration and isolation, and efficient networking.