On-going researches

Combined with his knowledge of the mixed and digital hardware implementations, he will continue to researches about the pattern matching scheme using many-core system. In his dissertation, the parallel string matching scheme was proposed for reducing hardware cost. Even though the string pattern matching schemes were studied for the many applications in the deep packet inspection, he thinks that the pattern matching scheme for more complex patterns should be adopted using many cores for the success of industrial fields. 

Prof. Kim started researches in the field of neural network implementation from his sabbatical year. He has interests in applying the approximate computing and stochastic computing to the neural network implementation. On the other hand, the application of neural networks is very interesting for him. 

Most of all, he and his advised students have studied the binarized neural networks (BNNs). As highly quantized neural networks, they have significant benefits in hardware resource and latency. Besides, BNNs can be a realistic form for implementing TinyML (Tiny Machine Learning), which can be extended into all MCU-based applications. Moreover, BNNs can be implemented as PIM (Processing In Memory). However, its degraded performance and real implementation with small MCUs are very challenging. We have focused on the training method for BNNs and platform developments for MCUs at this time. 

On-going Researches

 - Neural Network Implementation and its Application

 - Pattern Matching Scheme

 - Binarzied Neural Networks and Processing in Memory (To be updated!!)

- New Arithmetic and their Supports (To be updated!!)