ASCENT: Ferroelectric-based Compute-in-Memory Dynamical Engine (Ferro-CoDE) to Solve Hard Combinatorial Optimization

ASCENT：基于铁电的内存计算动态引擎 (Ferro-CoDE) 解决硬组合优化问题

• 批准号: 2132918
• 负责人: Nikhil Shukla
• 金额: $ 149.87万
• 依托单位: University of Virginia Main Campus
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-08-15 至 2025-07-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2132918&HistoricalAwards=false
• 关键词: ASCENT; Ferroelectric; based; Compute; Memory

Not all computing problems are created equal. At the heart of many, increasingly important, applications ranging from the design of intelligent machines that can explain their decisions, to electronic design automation for tamper-proof integrated circuits, lies a class of combinatorial optimization problems that remain an unconquered bastion of traditional digital computing. As a case in point, solving the archetypal Boolean satisfiability problem, integral to many such applications, requires exponentially increasing energy and computation time that makes its practical deployment at large scales unfeasible. The proposed research aims to address this challenge through a cohesive across-the-stack effort that spans from the formulation of a physics-inspired computational model to designing a supporting hardware ecosystem that encompasses novel engineered materials and devices, circuits, and their subsequent system integration. The foundational principle of the Ferroelectric-based Compute-in-Memory Dynamical Engine (FerroCoDE) is based on exploiting the inherent efficiency of physical phenomena in nature (namely, maximization of entropy production) by creating unique synergies between physics and computation. To execute these computational models efficiently, the team of researchers is developing a new hardware platform that converges the unique capabilities of dynamical systems with compute-in-memory architectures, enabled through fundamental innovation in ferroelectric materials and their device functionalities. The FerroCoDE platform will enable orders-of-magnitude improvement in computational efficiency enabling the deployment of relevant applications at a scale and in (energy-constrained) environments that are presently challenging to achieve using present day computers. Furthermore, to broaden the impact of this work, the team will develop a publicly accessible online platform, OscWorks, for applying oscillator-based computing in education and research. Additionally, the team will engage with K-12 and undergraduate students through various initiatives such as workshops, online seminars, and research opportunities.The systems challenge that is being addressed by this ASCENT project is to design, fabricate and demonstrate a Ferroelectric Hafnium Oxide (HfO2) based Compute-in-Memory (CiM) Dynamical Engine (FerroCoDE) that leverages the rich non-linear analog dynamics of oscillators in conjunction with the area and energy efficiency of ferroelectric CiM architecture to accelerate the computationally hard maximum satisfiability problem. The FerroCoDE exploits a novel formulation of the satisfiability problem as the direct maximization of entropy in the compute engine which is being developed through a vertically integrated materials-to-systems effort that focusses on: (i) Development of phase- and crystallographic-texture-engineered HfO2-based ultra-thin ferroelectric and antiferroelectric films with tunable properties; (ii) Design and fabrication of a novel non-volatile 1FeFET-1FTJ memory cell and array to enable in-memory programming and evaluation of the satisfiability clauses; and energy efficient AFE oscillator arrays (iii) Building synergistic convergence between the hardware and algorithm; (iv) System engineering, with emphasis on developing learning algorithms to optimize dynamical system initialization, development of annealing schedules and hardware scalability. (v) Development and demonstration of a FerroCoDE prototype on PCB.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

并不是所有的计算问题都是一样的。从能够解释其决策的智能机器的设计，到防篡改集成电路的电子设计自动化，在许多越来越重要的应用中，都存在一类组合优化问题，这些问题仍然是传统数字计算的一个未攻克的堡垒。作为一个恰当的例子，解决原型布尔可满足性问题（这是许多此类应用的组成部分）需要指数增长的能量和计算时间，这使得其在大规模的实际部署不可行。拟议的研究旨在通过跨堆栈的内聚性努力来解决这一挑战，从物理启发的计算模型的制定到设计支持硬件生态系统，包括新型工程材料和器件、电路及其后续系统集成。基于铁电的内存计算动力引擎（FerroCoDE）的基本原理是通过在物理和计算之间创造独特的协同作用，利用自然界中物理现象的固有效率（即熵产最大化）。为了有效地执行这些计算模型，研究团队正在开发一种新的硬件平台，该平台通过铁电材料及其设备功能的基本创新，将动态系统的独特功能与内存中计算架构融合在一起。FerroCoDE平台将使计算效率得到数量级的提高，从而能够在（能源受限的）环境中大规模部署相关应用程序，这是目前使用现有计算机难以实现的。此外，为了扩大这项工作的影响，该团队将开发一个可公开访问的在线平台OscWorks，用于将基于振荡器的计算应用于教育和研究。此外，该团队还将通过各种活动，如研讨会、在线研讨会和研究机会，与K-12和本科生接触。ASCENT项目正在解决的系统挑战是设计、制造和演示基于铁电氧化铪（HfO2）的内存中计算（CiM）动力引擎（FerroCoDE），该引擎利用振荡器丰富的非线性模拟动力学，结合铁电CiM架构的面积和能源效率，加速计算困难的最大可满意度问题。FerroCoDE利用了可满足性问题的新公式，作为计算引擎中熵的直接最大化，该计算引擎正在通过垂直整合材料到系统的努力开发，其重点是：(i)开发具有可调性能的基于hfo2的超薄铁电和反铁电薄膜；（ii）设计和制造一种新的非易失性1FeFET-1FTJ存储单元和阵列，以实现内存编程和可满足性条款的评估；（三）在硬件和算法之间建立协同收敛；（iv）系统工程，重点是发展学习算法，以优化动态系统初始化，退火计划和硬件可扩展性的发展。在PCB上开发和演示铁码原型。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

A 2-Transistor-2-Capacitor Ferroelectric Edge Compute-in-Memory Scheme with Disturb-Free Inference and High Endurance

具有无干扰推理和高耐用性的 2 晶体管 2 电容器铁电边缘计算内存方案

• DOI: 10.1109/led.2023.3274362
• 发表时间: 2023
• 期刊: IEEE Electron Device Letters
• 影响因子: 4.9
• 作者: Ma, Xiaoyang;Deng, Shan;Wu, Juejian;Zhao, Zijian;Lehninger, David;Ali, Tarek;Seidel, Konrad;De, Sourav;He, Xiyu;Chen, Yiming
• 通讯作者: Chen, Yiming

Oscillator-based Dynamical Computing Platforms to Solve Combinatorial Optimization

基于振荡器的动态计算平台解决组合优化

• DOI: 10.1109/edtm53872.2022.9798043
• 发表时间: 2022
• 期刊: 2022 6th IEEE Electron Devices Technology & Manufacturing Conference (EDTM
• 作者: Mallick, Antik;Bashar, Mohammad Khairul;Shukla, Nikhil
• 通讯作者: Shukla, Nikhil

Computational Models Based on Synchronized Oscillators for Solving Combinatorial Optimization Problems

• DOI: 10.1103/physrevapplied.17.064064
• 发表时间: 2022-06-30
• 期刊: PHYSICAL REVIEW APPLIED
• 影响因子: 4.6
• 作者: Mallick, Antik;Bashar, Mohammad Khairul;Shukla, Nikhil
• 通讯作者: Shukla, Nikhil

FeFET-Based Logic-in-Memory Supporting SA-Free Write-Back and Fully Dynamic Access With Reduced Bitline Charging Activity and Recycled Bitline Charge

• DOI: 10.1109/tcsi.2023.3251961
• 发表时间: 2023-06
• 期刊: IEEE Transactions on Circuits and Systems I: Regular Papers
• 作者: Wenjun Tang;Ming-En Lee;Juejian Wu;Yixin Xu;Yao Yu;Yongpan Liu;Kai Ni;Yu Wang;Huazhong Yang;V. Narayanan;Xueqing Li

Dynamical System-Based Computational Models for Solving Combinatorial Optimization on Hypergraphs

• DOI: 10.1109/jxcdc.2023.3235113
• 发表时间: 2023-06-01
• 期刊: IEEE JOURNAL ON EXPLORATORY SOLID-STATE COMPUTATIONAL DEVICES AND CIRCUITS
• 影响因子: 2.4
• 作者: Bashar, Mohammad Khairul;Mallick, Antik;Shukla, Nikhil
• 通讯作者: Shukla, Nikhil