AI-driven Design of Analog Integrated Circuits for Ultrasound Applications

模拟前端在大多数超声硬件系统中是必不可少的，广泛应用于通信、工业检测、医疗诊断、机器人增强传感、外科手术工具等。尽管在SoC中模拟电路面积通常不到20%，但它所需的设计工作量可以超过80%。模拟电路和数字电路之间的生产率差距不断扩大。虽然多年来数字电路的设计自动化水平一直在稳步提高，但模拟电路的设计自动化水平仍然很低。模拟IC设计方法在过去40年中几乎没有变化：它仍然是一种缓慢的体验和试错驱动的手动过程。最近，开发新的人工智能技术来自动化模拟IC的设计开始引起人们的注意。格拉斯哥大学的人工智能驱动设计实验室是人工智能驱动模拟IC设计的先驱。作为第一个将人工智能技术引入模拟IC设计的方法(2008)，2021年提出了人工智能驱动的模拟IC尺寸调整方法ESSAB，该方法首先解决了行业级高性能模拟积木(考虑了全套“难学”的性能)。相比之下，ESSAB在短短几个小时内就超过了经验丰富的设计师的设计质量。然而，在ESSAB中还没有考虑工艺、电压和温度(PVT)的变化。在此基础上，本项目的目标是(1)使用ESSAB设计各种用于超声应用的模拟IC，并取得良好的结果。(2)明确ESSAB的利弊。(3)提出了考虑PVT变化的ESSAB-II，以获得稳健的设计，为工业应用做好准备。这个项目是跨学科的。预计该学生将对人工智能技术有很好的理解，并通过学习拥有实现人工智能算法的技能。然而，更多的重点放在使用ESSAB和可能的ESSAB-II来实际设计超声模拟IC。

The analog front end is essential in most ultrasonic hardware systems, which is widely used in communications, industrial inspection, medical diagnosis, robotically enhanced sensing, surgical tools, etc. Although the analog circuit area is usually less than 20% in an SoC, its required design efforts can be more than 80%. The productivity gap between analog and digital circuits keeps widening. While the design automation level for digital circuits has been increasing steadily over the years, the design automation level for analog circuits remains very low. The analog IC design methodology remains almost unchanged over the past four decades: it is still a slow experience and trial-and-error-driven manual process. Recently, developing novel AI techniques to automate the design of analog ICs starts to attract attention. The AI-driven design lab, University of Glasgow, is a pioneer in AI-driven analog IC design. Being the first few to introduce AI techniques to analog IC design (2008), the AI-driven analog IC sizing method, called ESSAB, was proposed in 2021, which firstly addressed industry-level high-performance analog building blocks (considering the full set of "hard to learn" performances). Through comparison, ESSAB surpasses experienced designers' design quality in only a few hours. However, process, voltage, and temperature (PVT) variations have not been considered in ESSAB yet. Built upon this, this project aims to (1) Design various analog ICs for ultrasound applications using ESSAB achieving promising results. (2) Identify the pros and cons of ESSAB. (3) Propose ESSAB-II considering PVT variations obtaining robust designs, which is ready for industry use. This project is interdisciplinary. The student is expected to have a good understanding of AI techniques and have skills to implement AI algorithms through the study. However, more emphasis locates on using ESSAB and possible ESSAB-II to design ultrasonic analog ICs practically.