Abstract
Precision in analog integrated circuit design is often compromised by process variations across silicon wafers. Common-Centroid (CC) layouts are employed to mitigate these effects by interweaving transistor units to align their geometric centers. However, finding an optimal CC layout involves navigating a complex space of conflicting objectives, including dispersion, routing complexity, and manufacturing effects. This paper presents a unified multi-objective optimization framework using Archived Multi-Objective Simulated Annealing (AMOSA) to automate CC placement. By integrating topology exploration with advanced cost modeling for diffusion sharing and layout-dependent effects, the proposed tool identifies high-performance layouts that surpass existing manual and automated techniques in both matching accuracy and area efficiency.
Introduction
In analog chip manufacturing, "process variations" can lead to significant performance discrepancies between identical transistors. To counteract linear gradients in silicon properties, designers utilize Common-Centroid (CC) layouts. These layouts decompose transistors into smaller units and arrange them such that their centroids overlap. Despite its benefits, CC placement is a non-trivial task due to the "balancing act" required between several competing constraints:
- Dispersion: Maximizing unit spread to improve matching.
- Routing Complexity: Minimizing wire length and parasitic resistance.
- Diffusion Sharing: Merging active regions to reduce layout area.
- Layout Dependent Effects (LDE): Mitigating performance degradation from proximity to structures like well edges.
Existing automation tools often address these constraints in isolation or require extensive manual intervention. This research introduces a framework designed to optimize these objectives simultaneously.
Methodology
The core of the proposed framework is the AMOSA (Archived Multi-Objective Simulated Annealing) algorithm. Unlike single-objective optimization, AMOSA explores a "Pareto front"—a set of optimal solutions where improving one metric necessitates a trade-off in another.
Topology Exploration and Transformation
A key innovation is the use of geometric transformations, specifically XX/180° and XY/180°, to explore a vast space of valid CC patterns. This allows the tool to discover non-obvious topologies that offer superior trade-offs between routing and matching compared to traditional fixed patterns.
Diffusion Break and Dummy Management
The framework intelligently manages "diffusion breaks"—interruptions in the active silicon region—by placing dummy transistors. While dummies improve symmetry, they increase area and routing length. The optimizer treats dummy count as a primary objective, seeking "sweet spots" that balance area efficiency with performance.
Integrated Cost Modeling
The optimization loop incorporates direct calculations for:
- Degree of Dispersion: Quantifying the spatial spread of transistor units.
- Routing Cost: Estimated using Minimum Rectilinear Steiner Trees.
- LDE: Specifically modeling Well Proximity Effects to avoid "legal but bad" layouts.
Experimental Results
The tool was evaluated on industry-standard analog blocks, including Current Mirrors (CM), Cascode Differential Input Pairs (CDIP), and a full 5T-Operational Transconductance Amplifier (OTA).
Performance Benchmarks
- Mismatch Reduction: In current mirror tests, the proposed method reduced current mismatch to 0.1%, compared to 3.72% achieved by previous state-of-the-art methods.
- Area and Routing Efficiency: For CDLP circuits, the tool reduced the dummy count from 12 to 4, resulting in a 20% area reduction. Routing costs were consistently lowered by 5-15%.
- System-Level Impact: Post-layout simulations of a 5T-OTA showed that the tool's optimized topology achieved an offset voltage of 3.1 mV, a significant improvement over the 12.02 mV offset produced by standard topologies.
Conclusion
The multi-objective optimization framework successfully automates the complex task of CC placement for analog transistors. By considering LDE and diffusion sharing within the optimization loop, the tool generates layouts that are not only theoretically symmetric but also robust to real-world manufacturing effects. Future work will extend this framework to include passive components like resistors and capacitors and explore methods to optimize for random mismatch alongside systematic variations.
Summary updated