One failed memory cell out of millions cells will cause the whole memory circuit to fail without ECC (error checking and correction) techniques. That is why memory designers have high parametric yield requirements for the SRAM core cell. It requires no fails in hundreds of millions or billions of brutal force Monte Carlo simulations if foundry statistical models are accurate up to the high sigma region.Memory circuit designers also have high yield requirements for other circuit block and memory partitions, such as sense amplifier sor critical-path partitions.
How to analyze high yield and debug, improve the design in Virtuoso ADE
Virtuoso Analog Design Environment GXL selects WCD (worst case distance) metric-based method as its high yield solution. To estimate high yield, ADE GXL will first find the WCD point in the statistical space.Once the WCD point is found, the yield can be calculated directly using the WCD.
The accuracy of WCD is impacted by nonlinearity of spec boundary in statistical spaces. However, that error is not significant in most high yield applications. The WCD point has the shortest distance from nominal point to fail region in statistical space. It is also the most probable point to fail in statistical space. During a continuously running Monte Carlo process, the first failing Monte Carlo point has a very high probability to be very close to the WCD point in statistical space.This makes the WCD point very attractive for creating a high-yield statistical corner because it captures the circuit condition with the highest probability to fail.This point becomes increasingly important if designers want to debug and improve the design. Based on the WCD point, Virtuoso ADE GXL provides the capability to create a high-sigma corner to improve the yield.
The completed high yield solution in Virtuoso ADE GXL is a part of the TSMC AMS reference flow. It includes:
1. High yield calculation
2. Creation of high yield statistical corner
3. Optimization of the design with high yield corner
4. Verification of design using high yield calculation
What is coming next? New algorithms for high dimension!
Cadence R&D co-invented the next-generation high yield estimation algorithm with researchers from Carnegie Mellon University recently[1]. The scaled-sigma sampling algorithm works well with high-dimensional nonlinear problems, which exist in large circuit blocks. Please stay tuned!. The algorithm will be publicly available in the coming IC616ISR release.
[1] Shupeng Sun, Xin Li, Hongzhou Liu, Kangsheng Luo and Ben Gu, "Fast statistical analysis of rare circuit failure events via scaled-sigma sampling for high-dimensional variation space," IEEE/ACM International Conference on Computer-Aided Design (ICCAD), pp. 478-485, 2013.