WITINMEM

Founded in October 2017, WITINMEM (Beijing Zhicun Technology Co., Ltd.) is a semiconductor company focused on developing processing-in-memory (PIM) chips. The company was established by CEO Wang Shaodi and Co-founder & Chief Scientist Guo Xinjie. Both founders are alumni of Peking University's School of Electronics and went on to receive their PhDs in the United States, with Wang attending UCLA and Guo attending UCSB. Their academic research, which began as early as 2012, culminated in the successful development of the first analog PIM deep learning chip in 2016, laying the groundwork for the company's formation.

WITINMEM addresses the "storage wall" bottleneck of traditional Von Neumann computing architectures, where the constant transfer of data between memory and processing units consumes significant power and time. The company's core technology utilizes Flash memory to perform neural network computations directly where the data is stored, which can enhance computing efficiency by a factor of 10 to 50 compared to conventional deep learning chips. This business model centers on the design and sale of these high-efficiency, low-power AI chips to manufacturers of consumer electronics and other smart devices. By selling its chips and providing related development support, the company generates revenue from clients in sectors like wearable technology, smart home, and automotive.

The company has achieved significant milestones, including the mass production of its first PIM accelerator, the WTM1001, and its first PIM System-on-Chip (SoC), the WTM2101, which launched in 2022. The WTM2101 is an ultra-low-power chip designed for edge AI applications, enabling features like speech recognition, voice enhancement, and health monitoring in devices such as TWS earphones, hearing aids, and wearables with minimal power consumption. Another key product line, the WTM-8 series, is designed for higher-demand applications like real-time 4K-8K video processing. To support developers, WITINMEM provides a proprietary software stack, including the WITIN Mapper compiler and development tools, to facilitate the deployment of neural networks onto its chips.

Keywords: processing-in-memory, PIM, AI chips, edge computing, SoC, neural networks, low-power AI, semiconductor, computing architecture, flash memory