In physical design, human designers typically place macros via trial and error, which is a Markov decision process. Reinforcement learning (RL) methods have demonstrated superhuman performance on the macro placement. In this paper, we propose an extension to this prior work [1]. We first describe the details of the policy and value network architecture. We replace the force-directed method with DREAMPlace for placing standard cells in the RL environment. We also compare our improved method with other academic placers on public benchmarks.

No abstract

Abstract-The signal via is a heavily utilized interconnection structure in high-density System-on-Package (SoP) substrates and printed circuit boards (PCBs). Vias facilitate complicated routings in these multilayer structures. Significant simultaneous switching noise (SSN) coupling occurs through the signal via transition when the signal via suffers return current interruption caused by reference plane exchange. The coupled SSN decreases noise and timing margins of digital and analog circuits, resulting in reduction of achievable jitter performance, bit error ratio (BER), and system reliability. We introduce a modeling method to estimate SSN coupling based on a balanced transmission line matrix (TLM) method. The proposed modeling method is successfully verified by a series of time-domain and frequency-domain measurements of several via transition structures. First, it is clearly verified that SSN coupling causes considerable clock waveform distortion, increases jitter and noise, and reduces margins in pseudorandom bit sequence (PRBS) eye patterns. We also note that the major frequency spectrum component of the coupled noise is one of the plane pair resonance frequencies in the PCB power/ground pair. Furthermore, we demonstrate that the amount of SSN noise coupling is strongly dependent not only on the position of the signal via, but also on the layer configuration of the multilayer PCB. Finally, we have successfully proposed and confirmed a design methodology to minimize the SSN coupling based on an optimal via positioning approach.Index Terms-Noise coupling, power/ground noise, printed circuit board (PCB), reference plane change, simultaneous switching noise (SSN), system-on-package (SoP), via.

No abstract