A scalable σ-space based methodology for modeling process parameter variations in analog circuits

This paper describes a scalable method (called ALAMO) for modeling analog circuit performance in the presence of MOSFET process parameter variations, including (i) process/model parameter variation of individual devices and (ii) mismatch of process/model parameters of device pairs. This capability is very important for many popular analog circuits, such as current biasing circuits, voltage reference circuits, single-ended output amplifiers, and transconductance amplifiers. The methodology uses two steps, layout regularity analysis (LRA) and principal equation analysis (PEA), to reduce the complexity of the modeling problem. LRA finds devices and device pairs with similar statistical behavior. PEA eliminates less important correlations using a sensitivity-based metric. Experiments showed that ALAMO method is more accurate than Monte Carlo analysis and σ-space method, and scales well even for analog circuits involving hundreds of fingered MOSFETs. LRA and PEA are effective in decreasing the modeling problem complexity reducing the modeling time by more than 70% without significantly influencing accuracy.