Targeted system optimization using inverse design, Prerit Terway, G4 (3958781)

Most systems have multiple objectives and constraints. For example, a company making a smartphone needs to increase the battery life, minimize the lag, and maximize the bandwidth. Solutions to these design problems are given by the best tradeoff among the different objectives. Many recent works assign weights to the objectives to determine their importance. Assigning weights requires domain expertise or a trial-and-error method that leads to significant cost increases and larger turnaround times. INFORM (our method) uses AI to augment human creativity when designing a system with multiple objectives and constraints. Rather than wasting design time and computational resources to discover the best tradeoff among the different objectives, INFORM allows focusing only on the region of interest to the designer, thus speeding up the design process. In contrast to existing methods that use forward model to optimize system, we use inverse design. Inverse design utilizes the desired system response to perform targeted optimization. INFORM can drive an existing solution to another solution with different specifications. Driving a solution to another specification allows a company to have a generic system prototype and adapt it using INFORM for various customer needs. Many existing methods require starting optimization from scratch, thus increasing the design cost. INFORM does not use GPUs, thus presenting a sustainable design technique. Instead, we utilize multiple CPU cores to speed up the optimization. INFORM reduces synthesis time by up to 17x and improves the value of the objective function by up to 30% compared to state-of-the-art method.