带有铁电模拟内容的深度随机森林可定位存储器
Xunzhao Yin1,2, Franz Müller3, Ann Franchesca Laguna4
1Zhejiang University, Hangzhou, Zhejiang, China.
Science advances
|June 5, 2024
概括
这项研究介绍了一种使用铁电模拟内容可定位存储器 (ACAM) 的新型深度随机森林 (DRF) 加速器,以实现高效的边缘智能. 与现有硬件相比,FeFET ACAM DRF显著提高了能源效率,并减少了延迟.
科学领域:
- 人工智能的人工智能
- 计算机工程 计算机工程
- 材料科学 材料科学 材料科学
背景情况:
- 深度随机森林 (DRF) 模型为边缘情报任务提供了竞争力的准确性和可解释性,类似于深度神经网络 (DNN).
- 现有的DRF硬件加速器落后于DNN同行,特别是在决策节点高效处理关键的分支分割操作方面.
- 铁电材料为新的内存和计算架构提供了独特的特性.
研究的目的:
- 为深度随机森林 (DRF) 算法提出和演示一个高效的硬件加速器.
- 利用铁电模拟内容可定位存储器 (ACAM) 加快DRF的决策过程.
- 评估拟议的DRF加速器架构的性能,能源效率和可扩展性.
主要方法:
- 通过在铁电模拟内容可定位存储器 (ACAM) 细胞中使用关联搜索来实现DRF.
- 一个超紧的ACAM电池的设计,使用两个铁电场效应晶体管 (FeFET) 来存储决策边界作为模拟极化状态.
- 开发一个DRF加速器架构和用于将DRF模型映射到ACAM阵列上的方法.
- 基于FeFET ACAM DRF的功能,特性,可扩展性和稳定性与设备非理想性的实验和模拟验证.
主要成果:
- 拟议的FeFET ACAM DRF加速器在能源效率 (∼10^6×/10×) 和延迟 (∼10^6×/2.5×) 方面取得了显著的改进,与基于CPU和ReRAM的最新DRF硬件实现相比.
- 通过使用FeFETs中的模拟极化状态,ACAM单元使节能分支分裂操作成为可能.
- 该架构证明了对FeFET设备非理想性的稳定性,确保了可靠的操作.
结论:
- 铁电ACAM为深度随机森林 (DRF) 算法实现超紧和节能硬件加速器提供了一个有前途的途径.
- 拟议的FeFET ACAM DRF架构提供了实质性的性能提升,解决了边缘智能DRF硬件加速当前的局限性.
- 这项工作为下一代边缘人工智能系统铺平了道路,这些系统具有增强的计算能力和降低的功耗.
更多相关视频
09:49In Situ Transmission Electron Microscopy with Biasing and Fabrication of Asymmetric Crossbars Based on Mixed-Phased a-VOx
Published on: May 13, 2020
4.1K
10:05In Vitro Multiparametric Cellular Analysis by Micro Organic Charge-modulated Field-effect Transistor Arrays
Published on: September 20, 2021
2.4K
相关概念视频
01:18Capacitor With A Dielectric
Parallel plate capacitors consist of two conducting plates separated by a certain distance. However, it is mechanically difficult to hold the large plates parallel to each other without actual contact. Hence, a dielectric layer is commonly placed between the plates, which provides an easy solution for holding the plates together with a small gap and increases the capacitance of the capacitor.
Dielectrics are non-conducting materials with no free or loosely bound electrons. When a dielectric is...
Dielectrics are non-conducting materials with no free or loosely bound electrons. When a dielectric is...
01:31Ferromagnetism
Materials like iron, nickel, and cobalt consist of magnetic domains, within which the magnetic dipoles are arranged parallel to each other. The magnetic dipoles are rigidly aligned in the same direction within a domain by quantum mechanical coupling among the atoms. This coupling is so strong that even thermal agitation at room temperature cannot break it. The result is that each domain has a net dipole moment. However, some materials have weaker coupling, and are ferromagnetic at lower...
01:25MOS Capacitor
A Metal-Oxide-Semiconductor (MOS) capacitor is a fundamental structure used extensively in semiconductor device technology, particularly in the fabrication of integrated circuits and MOSFETs (metal-oxide-semiconductor field-effect transistors). The MOS capacitor consists of three layers: a metal gate, a dielectric oxide, and a semiconductor substrate.
The metal gate is typically made from highly conductive materials such as aluminum or polysilicon. Beneath the metal gate lies a thin layer of...
The metal gate is typically made from highly conductive materials such as aluminum or polysilicon. Beneath the metal gate lies a thin layer of...