Measurement of worst-case power delivery noise and construction of worst case current for graphics core simulation

Abstract

Worst case graphics core power delivery noise is a major indicator of graphics chip performance. The design of good graphics core power delivery network (PDN) is technically difficult because it is not easy to predict a worst case current stimulus during pre-silicon design stage. Many times, the worst case power delivery noise is observed when graphics benchmark software is run during post-silicon validation. At times like this, it is too late to rectify the power delivery noise issue unless many extra capacitor placeholders are placed during early design stage. To intelligently optimize the graphics core power delivery network design and determining the right amount of decoupling capacitors, this paper suggests an approach that setup a working platform to capture the worst case power delivery noise; and later re-construct the worst case power delivery current using Thevenin's Theorem. The measurement is based on actual gaming application instead of engineering a special stimulus that is generated thru millions of logic test-vectors. This approach is practical, direct and quick, and does not need huge computing resources; or technically skilled logic designers to design algorithms to build the stimulus.