Hardware-corroborated Variability-Aware SRAM Methodology

The paper presents a novel hardware-measurement based methodology to extract within-product-array variability as opposed to kerf. The methodology was applied to a 45nm 2.4Mb SRAM design with emphasis on critical dimension measurements. Local mismatch and spatial variations, LER effects and backend effects were measured. Lithographic models were tuned to the 45nm process. Statistically measured variability in critical process units along with extracted long-range CD variation through lithography simulations were then fed to a newly developed statistical engine to predict design yield, leakage and performance. The predicted yield results were then corroborated with the hardware providing feedback for process tuning. In addition to monitoring mature processes, such hardware measurement techniques can also be used to monitor or decipher yield-detractor defects.