Uploaded May 2026 | Updated September 2026, 2 weeks ago
Analog integrated circuit layout remains intensely manual, requiring weeks of expert effort. We present CLARA's multi-objective reward function that decomposes layout quality into interpretable components: symmetry, compactness, connectivity, and device grouping. Unlike sparse terminal rewards, our graduated penalty system provides dense feedback that scales proportionally to constraint violations. Validation across three SKY130 circuits demonstrates convergence from severely constraint-violating states to fully feasible layouts, with device grouping driving optimization while revealing fundamental trade-offs between matching and geometric efficiency.
Analog integrated circuit layout remains intensely manual, requiring weeks of expert effort. We present CLARA's multi-objective reward function that decomposes layout quality into interpretable components: symmetry, compactness, connectivity, and device grouping. Unlike sparse terminal rewards, our graduated penalty system provides dense feedback that scales proportionally to constraint violations. Validation across three SKY130 circuits demonstrates convergence from severely constraint-violating states to fully feasible layouts, with device grouping driving optimization while revealing fundamental trade-offs between matching and geometric efficiency.
