[论文解读] Device-Circuit-Architecture Co-Exploration for Computing-in-Memory Neural Accelerators
该论文提出NACIM,一种用于计算内存(CiM)神经加速器的跨层协同探索框架,可联合优化器件类型、电路拓扑和神经架构,同时考虑器件非理想性。通过整合感知非理想性的训练与多目标优化,NACIM在器件非理想性下仅造成0.45%的精度损失,显著优于最先进神经 Architecture Search(NAS)的76.44%损失,同时实现16.3 TOPs/W的能效,较先前工作提升3.17倍。
Co-exploration of neural architectures and hardware design is promising to simultaneously optimize network accuracy and hardware efficiency. However, state-of-the-art neural architecture search algorithms for the co-exploration are dedicated for the conventional von-neumann computing architecture, whose performance is heavily limited by the well-known memory wall. In this paper, we are the first to bring the computing-in-memory architecture, which can easily transcend the memory wall, to interplay with the neural architecture search, aiming to find the most efficient neural architectures with high network accuracy and maximized hardware efficiency. Such a novel combination makes opportunities to boost performance, but also brings a bunch of challenges. The design space spans across multiple layers from device type, circuit topology to neural architecture. In addition, the performance may degrade in the presence of device variation. To address these challenges, we propose a cross-layer exploration framework, namely NACIM, which jointly explores device, circuit and architecture design space and takes device variation into consideration to find the most robust neural architectures. Experimental results demonstrate that NACIM can find the robust neural network with 0.45% accuracy loss in the presence of device variation, compared with a 76.44% loss from the state-of-the-art NAS without consideration of variation; in addition, NACIM achieves an energy efficiency up to 16.3 TOPs/W, 3.17X higher than the state-of-the-art NAS.
研究动机与目标
- 通过利用计算内存(CiM)突破存储墙,解决冯·诺依曼架构的性能瓶颈。
- 实现器件、电路与神经架构设计层级的联合优化,以实现最高的硬件效率与网络精度。
- 缓解新兴非易失性存储器件中器件非理想性对神经网络推理精度的影响。
- 开发一种可扩展的多目标协同探索框架,同时优化精度、能耗、延迟、面积与硬件寿命。
- 证明结合非理想性感知训练的架构搜索,相比固定架构方法,可产生更鲁棒且高效的神经加速器。
提出的方法
- NACIM采用多目标强化学习框架,联合探索器件类型、电路拓扑与神经架构设计空间。
- 集成一种非理想性感知的训练流程,在反向传播中模拟器件非理想性,以提升鲁棒性。
- 框架使用包含归一化硬件效率(按延迟、面积与能耗加权)与推理精度的奖励函数。
- 采用蒙特卡洛采样估算器件非理想性在训练期间的影响,确保在器件参数分布下的鲁棒性。
- 搜索空间包含多种神经架构,如VGG-11、残差网络(ResNet)类结构及U-Net风格编码器,以探索高效、任务特定的设计。
- 硬件指标如EDP与面积经归一化后,作为硬件效率目标的一部分,以实现帕累托最优的权衡。
实验结果
研究问题
- RQ1在CiM加速器中,对器件、电路与神经架构设计空间的协同探索,是否能比传统NAS带来更高的精度与硬件效率?
- RQ2器件非理想性如何影响映射到CiM硬件的神经网络的推理精度?该影响是否可在训练阶段缓解?
- RQ3架构搜索对CiM神经加速器在非理想器件行为下的鲁棒性与效率有何影响?
- RQ4统一框架能否同时优化CiM加速器的精度、能耗、延迟、面积与硬件寿命?
- RQ5所提出的NACIM框架在不同主干网络架构与机器学习任务上的可扩展性如何?
主要发现
- NACIM实现16.3 TOPs/W的能效,相比最先进NAS方法提升3.17倍。
- 在器件非理想性下,NACIM仅造成0.45%的精度损失,远优于非感知非理想性的NAS方法的76.44%损失,展现出显著鲁棒性。
- 在CIFAR-10数据集上,采用ResNet类主干网络时,NACIM将精度从基线的72.18%提升至73.45%,同时面积减少23.3%,EDP降低52.4%。
- 在VLS目标分割任务中,NACIM实现93.12%的精度,面积为4.75×10⁸ μm²,EDP为5.54×10¹⁵ pJ·ns,精度与效率均优于基线。
- 该框架成功识别出更小、更高效的架构(如RLS),其精度更高、EDP更低,优于VGG-11等大型主干网络。
- 非理想性感知训练显著提升推理鲁棒性,相比非感知训练,RUL、VLS与EDS任务的精度分别提升9.8%、9.6%,IOU分别提升0.525。
更好的研究,从现在开始
从阅读论文到最终审阅,大幅缩短您的研究时间。
无需绑定信用卡
本解读由 AI 生成,并经人工编辑审核。