Dennard scaling
In plain wordsThe rule that as transistors shrank they also used proportionally less power, so chips got faster without getting hotter. It stopped working in the mid-2000s, ending the era of ever-faster single processors.
The scaling rule for MOSFET transistors stating that if all dimensions and the supply voltage are reduced by the same factor k, transistor density rises by k², switching speed by about k, and power per transistor falls by k², so power density stays constant. Together with Moore's lawMoore's law, n.: Gordon Moore's prediction that chipmakers could pack twice as many transistors onto a chip roughly every two years — a pace that for decades made computers steadily faster and cheaper, and has slowed but not stopped. it delivered three decades of faster chips that used less energy per operation, with steadily rising clock frequencies. It broke down around 2005–2007 because voltages could no longer fall: lowering the threshold voltage further caused unacceptable leakage current. Consequences included the stall of CPU clock speeds at a few gigahertz, the shift to multicore and massively parallel processors (favoring GPUsGPU, abbr.: A computer chip first designed to draw video-game graphics that turned out to be ideal for AI, because it can do thousands of simple calculations at the same time.), 'dark silicon' (parts of a chip that must stay unpowered at any moment) and the turn to specialized acceleratorsAI accelerator, n.: Any chip built or tuned to speed up AI — graphics chips, Google's TPUs, phone 'neural engines' and many startup designs. Such chips do the heavy math of training and running models. and lower-precision arithmetic for energy efficiency.
Origin
Described by Robert H. Dennard and IBM colleagues Fritz Gaensslen, Hwa-Nien Yu, V. Leo Rideout, Ernest Bassous and Andre LeBlanc in 'Design of ion-implanted MOSFET's with very small physical dimensions' (IEEE Journal of Solid-State Circuits, October 1974). The eponym is a later label, most often invoked in discussions of the rule's breakdown in the mid-2000s.
Related
- Moore's lawMoore's law, n.: Gordon Moore's prediction that chipmakers could pack twice as many transistors onto a chip roughly every two years — a pace that for decades made computers steadily faster and cheaper, and has slowed but not stopped.
- process nodeprocess node, n.: The label for a generation of chip-making technology, like '3 nanometer'. Smaller numbers mean newer, denser chips, but the number no longer measures any actual part of the chip.
- DRAMDRAM, abbr.: The main working memory of computers and AI chips: dense and fairly cheap, but slower than on-chip memory and needing constant refreshing to keep its data.
- GPUGPU, abbr.: A computer chip first designed to draw video-game graphics that turned out to be ideal for AI, because it can do thousands of simple calculations at the same time.
- AI acceleratorAI accelerator, n.: Any chip built or tuned to speed up AI — graphics chips, Google's TPUs, phone 'neural engines' and many startup designs. Such chips do the heavy math of training and running models.
Sources
- Dennard, R. H., Gaensslen, F. H., Yu, H.-N., Rideout, V. L., Bassous, E. & LeBlanc, A. R. (1974). Design of ion-implanted MOSFET's with very small physical dimensions. IEEE Journal of Solid-State Circuits SC-9(5).
Mentioned in
Moore's lawMoore's law, n.: Gordon Moore's prediction that chipmakers could pack twice as many transistors onto a chip roughly every two years — a pace that for decades made computers steadily faster and cheaper, and has slowed but not stopped., process nodeprocess node, n.: The label for a generation of chip-making technology, like '3 nanometer'. Smaller numbers mean newer, denser chips, but the number no longer measures any actual part of the chip., DRAMDRAM, abbr.: The main working memory of computers and AI chips: dense and fairly cheap, but slower than on-chip memory and needing constant refreshing to keep its data.