On the nature and effect of power distribution noise in CMOS digital integrated circuits

The thesis reports on the development of a novel simulation method aimed at modelling power distribution noise generated in digital CMOS integrated circuits. The simulation method has resulted in new information concerning: 1. The magnitude and nature of the power distribution noise and its dependence on the performance and electrical characteristics of the packaged integrated circuit. Emphasis is laid on the effects of resistive, capacitative and inductive elements associated with the packaged circuit. 2. Power distribution noise associated with a generic systolic array circuit comprising 1,020,000 transistors, of which 510,000 are synchronously active. The circuit is configured as a linear array which, if fabricated using two-micron bulk CMOS technology, would be over eight centimetres long and three millimetres wide. In principle, the array will perform 1.5 x 10 to the power of 11 operations per second. 3. Power distribution noise associated with a non-array-based signal processor which, if fabricated in 2-micron bulk CMOS technology, would occupy 6.7 sq. cm. The circuit contains about 900,000 transistors, of which 600,000 are functional and about 300,000 are used for yield enhancement. The processor uses the RADIX-2 algorithm and is designed to achieve 2 x 10 to the power of 8 floating point operations per second. 4. The extent to which power distribution noise limits the level of integration and/ or performance of such circuits using standard and non-standard fabrication and packaging technology. 5. The extent to which the predicted power distribution noise levels affect circuit susceptibility to transient latch-up and electromigration. It concludes the nature of CMOS digital integrated circuit power distribution noise and recommends ways in which it may be minimised. It outlines an approach aimed at mechanising the developed simulation methodology so that the performance of power distribution networks may more routinely be assessed. Finally. it questions the long term suitability of mainly digital techniques for signal processing.