TetraMem

TetraMem Inc. is a semiconductor company founded in 2018 by Glenn Ge, Miao Hu, Qiangfei Xia, and Joshua Yang, a team with shared experience in memristor research at Hewlett Packard Labs. The founders identified the unsustainable energy consumption of AI as a major obstacle and established the company to develop a new compute architecture that merges performance with sustainability. Their collective background in device modeling, process development, material science, and circuit design underpins the company's approach. CEO Glenn Ge departed a long career at HP, driven by the vision to advance efficient AI computing in the wake of significant AI advancements. Co-founders Qiangfei Xia and Joshua Yang serve as chief advisors, leveraging their extensive academic and research experience in memristive devices and neuromorphic computing.

Operating out of Silicon Valley, TetraMem focuses on in-memory computing (IMC) to address the hardware bottlenecks that emerge as AI algorithms become more complex. The company designs and develops neural processing units (NPUs) based on its proprietary analog in-memory computing technology. This technology is centered around a 'computing memristor', a type of resistive RAM (RRAM) that performs analog mathematical operations directly in memory, drastically reducing the energy-intensive process of shuttling data between memory and a processor. This approach is particularly suited for power-constrained edge and endpoint inference applications, such as in battery-powered mobile systems, robotic vision, security cameras, smart sensors, and automotive systems. The company’s business model appears to include IP licensing and the sale of its AI accelerator chips.

TetraMem's core product is an AI accelerator System-on-Chip (SoC) that utilizes its multi-level RRAM technology. This chip, named 'Cullinan', is engineered to be highly efficient, targeting a significant improvement in Tera Operations Per Second per Watt (TOPS/W) compared to conventional digital chips. The technology enables high-precision calculations within dense arrays of memristor devices, which provides a path to lower power consumption for neural network-based machine learning. Since its founding, the company has produced multiple test chips and is developing a commercial deep-learning accelerator. TetraMem serves clients such as system and product architects, ML engineers deploying AI in power-constrained devices, and software developers. The company has secured investments from firms including SAIC Capital and Foothill Ventures.

Keywords: in-memory computing, memristor, RRAM, AI accelerator, edge computing, neural processing unit, analog computing, semiconductor, deep-learning accelerator, low-power AI, neuromorphic computing, System-on-Chip, edge AI, AI hardware, energy-efficient computing, computing memristor, IoT sensors, machine learning hardware, sustainable AI, computer architecture, fabless, inference processing