With the development of applications such as generative artificial intelligence, computing power is becoming an increasingly important means of production. Our demand for computing power continues to rise. Simultaneously, the slowdown in Moore's Law indicates that performance improvements in traditional silicon-based semiconductor devices will not increase as quickly as in past decades. In addition, the physical separation of data and processing units in traditional computing architecture has become another major obstacle to the improvement of computing power, limiting the further improvement of computing efficiency. It is challenging to find efficient solutions to some complex problems under traditional computing paradigms.

In this ChalkTalk, we will explore new computing paradigms based on analog circuits in the context of the above issues. The analog computing paradigm can significantly reduce the need for precise arithmetic in specific applications through stochastic computation, thereby reducing the time and energy consumption required for calculations. In addition, the sensor-computing integrated circuit can significantly reduce data transmission time and improve energy efficiency by integrating data generation and processing units. Special-purpose accelerators based on analog circuits are more computationally efficient for a range of complex problems. Through this innovative computing paradigm, we are expected to solve the existing computing bottleneck problem and provide more powerful support for future applications such as artificial intelligence.


 Prof. Jing Jin

 School of Electronic Information and Electrical Engineering, SJTU


        2023.12.20 12:00-13:30