基于动态近似计算的近阈值数字电路设计技术研究

Maximizing the energy efficiency of a chip design has always been an important research topic, and challenged with the continuous development of semiconductor technologies, and their extensive applications in the fields of biomedical, big data processing, and mobile computing. Energy efficiency imposes a key constraint on the design. Approximate computing is an emerging design paradigm that exploits the intrinsic ability of applications to produce imprecise outputs. The cooperation with the near-threshold circuit design techniques provides a great opportunity to improve energy efficiency in error resilient applications. In this project, we shall begin with the analysis on the resilience of any given application, deriving the error performance model as well as the power-delay model. With the exploration on the tradeoffs among the key metrics, such as supply voltage, computation accuracy, circuit delay and power dissipation based on a variety of real-time input patterns, a comprehensive study on so called dynamic approximate computing shall be carried out. A systematic approach is also proposed to provide design automation. In the meanwhile, design and characterization techniques on standard cells from digital library as well as new SRAM cell are explored in near-threshold voltage regime. An energy efficient circuit design solution is presented with high reliability and noise margins. This study may enlighten the research in the future for the high energy-efficient digital circuits design in the nanometer scale era.

如何提高芯片的能量效率一直以来都是IC设计领域一个重要的课题及挑战，特别是随着大数据处理、移动计算、生物医疗等应用的兴起和普及，电路能效性更是成为影响芯片设计方案成功与否的关键。低压设计技术是实现高能效VLSI电路的一项关键技术，而近年来逐渐兴起并应用于硬件设计中的近似计算技术则为高能效设计另辟蹊径，两者的结合有望成为大幅度提高电路能效的一个重要途径。本课题拟从探究应用本身的弹性出发，根据对输入特性的概率分布分析，通过建立计算误差模型，探索电源电压、计算精度、电路延迟与功耗等关键参数之间相互制约与平衡的关系，提出一个考虑近似计算效率的电路能效优化方案，形成一种通用的、成体系的自动化设计方法。此外，通过对近阈值标准单元建库技术以及低压存储单元与阵列的研究与设计，解决近阈值条件下高稳定性、高能效电路的实现问题。本课题的研究旨在为纳米工艺下实现高能效数字电路提供一种较新的解决思路。

高能效芯片设计一直以来都是集成电路设计领域的一个重要课题及挑战，特别是近年来随着移动计算、能量采集、生物医疗等应用的兴起和普及，芯片的能效性成为直接影响其设计方案成功与否的关键。本项目研究基于动态近似计算的近阈值数字电路设计技术，从理论模型、设计方法、单元电路、功能模块，以及电路系统这几方面展开研究，对如何实现高能效数字集成电路设计提供一种可借鉴的解决思路。本项目主要完成的内容包括以下五个方面：1、完成适用于近阈值VLSI设计的数字标准单元库设计与验证；2、提出近阈值电压下的高稳定性、高能效数字电路设计技术与方法；3、提出一种基于近似计算的成体系的自动化设计流程和方法；4、完成高能效计算单元设计与实现；5、应用于图像处理的离散余弦变换器（DCT）系统设计与实现。本项目在实施过程中共计培养博士生1名，硕士生12名(包括1名转硕博连读)，发表学术论文10篇，包括6篇SCI论文和4篇IEEE/ACM国际会议论文，申请12项中国发明专利，其中3项已获授权，完成预期目标。