Edge De-Identification System Based on Computing in Memory Technology

Computing-in-Memory (CIM) is a highly energy-efficient architectural framework. By integrating privacy protection based on analog computing noise, it delivers an exceptional solution for de-identification. Combining CMOS circuit technology developed by ITRI’s Electronic and Optoelectronic System Research Laboratories, this technology creates a high-performance CIM chip achieving an extraordinary energy efficiency of over 5,000 TOPS/W (@4 bits). Furthermore, through the integration of the CIM chip with an FPGA system board, we have developed a series of high-performance neural network computing and cybersecurity applications.

This demonstration highlights a high-performance, edge-based de-identification image protection system. The system effectively addresses the long-standing challenge of using cameras for recording in sensitive areas without compromising individual privacy. Looking ahead, this system can be widely applied to home care, medical monitoring, and event detection devices, bringing greater convenience and privacy security to everyday life.

Ultimately, this technology not only showcases Taiwan's innovative capabilities in high-performance computing and low-power technology, but also drives the upgrade of domestic industries—strengthening Taiwan's competitiveness in the global tech market and unlocking new development opportunities.