Transimpedance amplifiers (TIA), as the critical optical receiver frontend to carry out the optical-to-electrical (O-E) conversion, plays a key role in these PNN-based systems. More specifically, TIA takes charge of converting the weak output current from the photodetector (PD) into voltage signals for further signal processing and computation in the electrical domain. To faithfully realize optical computations in PNNs, TIAs shall attain large bandwidth and high sensitivity. An enhanced RGC-based TIA with a cross-coupled pair and a series of variable gain amplifier (VGAs) is proposed and realized in this investigation using cost-effective 180-nm CMOS process to meet the requirements of analog-to-digital converter (ADC) voltage input range of the PNN system realization. The onsilicon measurement results show that the bandwidth = 1.5 GHz, the gain = 48.9 dBΩ, and 3.327 mW power dissipation, which is adequate to be integrated within PNNs without any loss of performance.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

A Transimpedance Amplifier Design with Tunable Gain and DC Offset for PNN-Based AI Edge Applications

  • Chua-Chin Wang,
  • Jyun-Wei Chen,
  • Yung-Jr Hung,
  • Zong-Ming Chang,
  • Chewn-Pu Jou,
  • Lan-Chou Cho

摘要

Transimpedance amplifiers (TIA), as the critical optical receiver frontend to carry out the optical-to-electrical (O-E) conversion, plays a key role in these PNN-based systems. More specifically, TIA takes charge of converting the weak output current from the photodetector (PD) into voltage signals for further signal processing and computation in the electrical domain. To faithfully realize optical computations in PNNs, TIAs shall attain large bandwidth and high sensitivity. An enhanced RGC-based TIA with a cross-coupled pair and a series of variable gain amplifier (VGAs) is proposed and realized in this investigation using cost-effective 180-nm CMOS process to meet the requirements of analog-to-digital converter (ADC) voltage input range of the PNN system realization. The onsilicon measurement results show that the bandwidth = 1.5 GHz, the gain = 48.9 dBΩ, and 3.327 mW power dissipation, which is adequate to be integrated within PNNs without any loss of performance.