基于强化图注意力网络的数字芯片布局方法  

Digital Chip Layout Method Based on Reinforcement Graph Attention Network

作　　者：侯泓秋 仝明磊 李易婉 HOU Hongqiu;TONG Minglei;LI Yiwan(College of Electronics and Information Engineering,Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]上海电力大学电子与信息工程学院,上海200090

出　　处：《计算机测量与控制》2024年第11期235-242,共8页Computer Measurement &Control

基　　金：国家自然科学基金项目(62105296)。

摘　　要：在数字芯片设计后端流程中,宏和标准单元的布局是一项耗时的工作,通过机器学习快速有效地提供解决方案能够加快芯片开发的周期,降低人工布局带来的风险;然而布局问题是一个多目标优化问题,目前大多数方法都注重在满足各项指标下最大化减小线长,已换取时钟延迟的降低,忽略了其他指标仍然存在下降的空间,例如良好的拥塞指标有利于降低芯片散热和功耗;针对上述问题,设计一种新的带有密集型奖励函数的深度强化学习框架,将拥塞信息映射到图像中,给出新的特征嵌入模型对版图的全局信息进行多尺度提取,并引入图注意力网络捕获网表的连接关系,采用Advantage Actor Critic(A2C)算法更新策略函数,实现了数字版图的自动布局,并在公共的数字芯片网表基准上验证了该方法的有效性。In the back-end process of digital chip design,it is a time-consuming task for the placement of macros and standard cells,a machine learning can provide a fast and effective solution,accelerating the cycle of chip development and reducing risks caused by manual layout.However,the layout is a multi-objective optimization problem.Currently,most methods focus on maximizing the reduction of line length when meeting various indicators,with an exchange for a decrease in clock delay,while ignoring the potential for further decline in other indicators,such as good congestion indicators that are benefical for reducing chip heat dissipation and power consumption;To solve the above problems,this paper proposes a new deep reinforcement learning framework with intensive reward function,maps the congestion information to the images,provides a new feature embedding model to extract the global information of the layout at multiple scales,introduces the connection relationship of the graph attention network to capture the netlist,updates the policy function by using the Advantage Actor Critic(A2C)algorithm,realizes the automatic layout of the digital landscape,and verifies the effectiveness of the proposed method on the public digital chip netlist benchmark.

关 键 词：图卷积神经网络 GAT 数字集成电路 深度强化学习 EDA 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]