Cupertino, CA, August 30, 2022 – Aizip, Inc., a Silicon Valley startup company providing leading artificial intelligence (AI) solutions for the Internet of Things (IoT), congratulated its founding member Weier Wan for publishing his monumental research work on Processing-In-Memory AI chip in Nature, a coveted leading scientific journal for breakthrough discoveries or innovations of significant impact to society.
Processing-in-memory (PIM) based artificial intelligence chip technology has gained significant traction for its potential to enable revolutionary AI applications in low-power edge/IoT devices. Its combined performance, efficiency, and cost advantages could bring complex AI functionalities into small battery-powered devices, including smart wearables, cameras, and industrial IoT sensors.
Weier Wan, leader of Aizip’s PIM program and a recent Ph.D. graduate from Stanford University, has published his research work on PIM in Nature. The article is titled “A compute-in-memory chip based on resistive random-access memory.” The paper has gained wide attention and has been covered by renown media across the globe, including IEEE Spectrum, The Sun, The Register, Independent, EE News, ScienceDaily, Guokr, and Tencent.
The paper describes a new PIM AI chip, named NeuRRAM, that simultaneously addresses three major challenges for edge AI hardware: efficiency, versatility, and accuracy. It supports a wide range of neural network models while consuming a small fraction of energy compared with today’s digital AI chips. It is the first large-scale hardware demonstration of RRAM-based PIM technology on a variety of AI applications.
The work is the result of a collaboration between the lab of H.-S. Philip Wong, a professor at Stanford University and an advisor of Aizip, and the lab of Gert Cauwenberghs, a professor at UC San Diego, along with several other research groups. Wan is the first author of the paper and led this research project.
“The conventional wisdom is that the higher efficiency of PIM is at the cost of versatility and accuracy,” said Cauwenberghs, who is a co-senior author of the paper. “With NeuRRAM, we showed that you don’t need to sacrifice any of them.” “The key to this breakthrough is the full-stack algorithm-hardware co-design,” added Wan.
Such full-stack co-design will play an even more important role in the transition of the PIM technology from research labs to practical use. Aizip has previously announced a full-stack design service covering the entire PIM product-development cycle, from Aizip’s industry-leading and production-grade tiny machine learning (TinyML) models to chip architecture design and silicon IPs. “By combining Aizip’s TinyML models and PIM-based AI chips, we strive to get the best of both worlds,” said Wan.
Aizip has been working with IC partners to commercialize the PIM technology. The design services can be tailored for specific customer needs to meet the low-power, low-cost, and accuracy requirements for a broad range of edge AI applications, including intelligent audio, vision, and time-series. The research work published in Nature by Wan is an excellent candidate for the next-generation PIM IC device.
With deep domain expertise covering the full stack of AI system design, from data and algorithm to embedded system and hardware design, Aizip is dedicated to providing best-in-class AI solutions for the Internet of Things. For more information, please visit www.aizip.ai.
* * *
About Aizip, Inc.
Aizip, Inc., develops AI models for edge applications. Based in Silicon Valley, Aizip provides models with superior performance, quick turnaround time, and excellent ROI. These models can be used in a wide range of applications for an intelligent, automated, and connected world. For more information, visit www.aizip.ai.