System and method for optimizing integrated circuit layout based on neural network

A system for optimizing an integrated circuit layout based on a neural network which interlocks with an electronic design automation (EDA) tool, the system comprising:

• a design information acquisition unit parsing a netlist file of a target integrated circuit and acquiring design information of the target integrated circuit;
• a circuit block decision unit deciding circuit blocks which correspond to a placement position adjustment target and each includes at least one element or cell based on the design information;
• a placement information generation unit generating first placement information indicating placement positions of the circuit blocks by using a neural network model learned to automatically place the circuit blocks based on a reinforcement learning algorithm;
• a placement information filtering unit predicting a minimum substrate area required for placing the circuit blocks according to the first placement information and when the predicted minimum substrate area does not correspond to a predetermined reference area range, discarding the first placement information and requesting the placement information generation unit to generate second placement information;
• an EDA tool control unit providing the first placement information to the EDA tool when the predicted minimum substrate area corresponds to the reference area range, and controlling the EDA tool to measure consumption power, performance, and an area of the target integrated circuit based on the first placement information;
• a first index generation unit generating a consumption power index, a performance index, and an area index based on measurement values for the consumption power, performance, and area of the target integrated circuit, which are measured through the EDA tool;
• and
• a reward providing unit calculating a reward corresponding to the first placement information based on a first reward component acquired by applying a first weight to the consumption power index, a second reward component acquired by applying a second weight to the performance index, and a third reward component acquired by applying a third weight to the area index, and providing the calculated reward to the neural network model.