Search

In this paper, we propose the through silicon via (TSV) array design optimization method using deep reinforcement learning (DRL) framework. The agent trained through the proposed method can provide an optimal TSV array that minimizes far-end crosstalk (FEXT) in one single step. We define the state, action, and reward that are elements of the Markov Decision Process (MDP) for optimizing the TSV array considering FEXT and train a deep q network (DQN) agent. For verification, we applied the proposed method to a 3 by 3 through silicon via array at stacked DRAM of High Bandwidth Memory (HBM). The network converged well, and as the result, the proposed method provided the optimal design that satisfies the target FEXT in which 3 dB lower than the initial design.

In this paper, we propose artificial intelligent (AI)router, a reinforcement learning (RL)-based auto-router considering signal integrity (SI), for the first time. Our algorithm has two main stages. At first, we design the transformer-based novel neural architecture considering the keep-out region, crosstalk region, and the number of vias for SI optimization. Then, the designed neural network is optimized by the policy gradient, one of the RL algorithms. Compared with the conventional maze routers, the A* algorithm, and the lee algorithm, it is verified that our AI-router outperforms the algorithms in terms of wire-length and crosstalk in a specific test case. Furthermore, it is shown that AI-router successfully performs multi-layer routing which is not feasible with conventional maze routers.