All objects surrounding us can be sensed so that we can obtain useful and various information such as voice, light, pressure, humidity, gas,  and temperature. The information is detected as electrical signals like voltage or current depending on sensors and converted to digital data by sensor interface circuits, which consist of front-end amplifiers and data converters. As a consequence, they have been widely used in many computing platforms. We focus on designing such energy-efficient interface circuits.

Rapid advances in semiconductor process scaling have enabled to realize miniaturized computing platforms implemented as the complete sensor systems. For internet-of-everything (IoE), since they can get much closer to sensing objects, the quality of the sensing signal will be greatly improved, and they will collect massive information in real life. Furthermore, the information can be used for AI applications such as feature extraction and classification. We focus on implementing the miniaturized sensor system including sensor interfaces, memory interfaces, microcontrollers, power management ICs, and clock management ICs.

Recently, growing efforts have been put in exploring human brain for brain-machine interface. Medical instruments are inherently rigid and cause side effects while interfacing with flexible body. The aforementioned miniaturized sensor system has a strong benefit to reduce risk of human implantation. In particular, the neural interface can read out the output signals (either voltage or capacitance) of the microelectrodes due to ionic concentration changes generated by neuron activity. In order not to damage brain tissues and to precisely detect the neural signals, we focus on designing ultra-low-power interface circuits.

Recently, modern neural networks inside the AI processors have excessive memory usage. The problem is not only in storage capacity of the memory, but also processing and transmission, since it costs a lot of energy and latency. Consequently, data needs to be processed and moved from the memory to the processing unit. The in-memory computing (IMC) is an emerging paradigm. The memory is allocated with internal computing capabilities. It enables minimizing data transfer between processor and memory. Especially, the analog IMC system stands out as a potentially more energy-efficient and compact alternative than digital counterparts for internet-of-things (IoT) applications. In the hardware, the multiply-and-accumulation (MAC) operation of the whole dataset and the weights accounts for the most power- and time-consuming computations in the networks. The digital input drives the synaptic weights through the input driver or the digital-to-analog converter (DAC). Then, each multiplied output located at a single column is accumulated in analog domain, which is converted to the digital MAC result by the output converter or the analog front-end (AFE) and the analog-to-digital converter (ADC). Notably, the energy efficiency is enhanced through the design of energy efficient ADC and DAC in various ways.