Dennard Scaling
"As transistors shrink, their power density holds constant, so smaller means faster for free."
"As transistors shrink, their power density holds constant, so smaller means faster for free."
Dennard scaling is the observation that as transistors get smaller, their power density stays roughly constant, so power consumption scales with area. As a transistor's dimensions shrink, its voltage and current scale down too, keeping power per unit area fixed while the smaller, faster transistors allow higher clock speeds. For decades it meant each new generation of smaller transistors could run faster within the same power budget on the same chip footprint.
Rated Empirical: grounded in studies or measurable evidence. The other 608.
Shrink a transistor's length, width, oxide thickness, and operating voltage all by the same factor, and the capacitance falls by that factor, the switching delay falls with it so the device is faster, and the power per transistor falls by the square of the factor. Meanwhile the number of transistors per unit area rises by the square. The two effects cancel in power terms: more transistors, each using proportionally less power, leaves the power per unit area constant. So a denser, faster chip did not run hotter: the benefit that, alongside Moore's law, drove decades of rising clock speeds.
The technical name for the idealised model, in which shrinking dimensions and voltage together keep the electric field constant and the power density fixed.
The companion observation about transistor count doubling, distinct from Dennard scaling's power-and-performance benefit; Moore's law continued after Dennard scaling broke.
It was the engine, alongside Moore's law, behind the exponential rise of computer clock speeds through the 1990s and early 2000s: the reason smaller transistors were worth building. Its breakdown around 2005 ended the era of ever-faster single processors and forced the whole industry toward multicore and parallelism, reshaping how software must be written to keep gaining speed.
Dennard scaling was an idealised approximation, the constant-field model, that ignored leakage current, which grew from negligible to dominant as transistors shrank. It broke down around 2005 to 2007: as devices reached very small sizes, subthreshold and gate leakage and threshold-voltage limits meant the operating voltage could no longer keep falling, so power density began to rise instead of holding constant. This ended frequency scaling and produced the power wall and dark silicon. It must be kept distinct from Moore's law, which concerns transistor count and continued for years after Dennard scaling failed.
Dennard scaling is not Moore's law — Moore's law is about transistor count and density, Dennard scaling about power density and performance per transistor, and confusing the two is the commonest error. Clock speeds did not stop rising because Moore's law ended; Moore's law continued, and it was Dennard scaling that broke. And smaller transistors do not automatically mean lower total chip power. That held only while Dennard scaling did, after which leakage made power density climb.
Yes, as far as the evidence goes. Dennard Scaling is rated Empirical here: it rests on studies or measurements rather than on a saying.
Rated from two sources. 609 of the 1,116 entries carry this rating. How the scale works.
The best-known laws are the least likely to rest on measurement.
Formulated by Robert Dennard and colleagues at IBM in their 1974 paper 'Design of Ion-Implanted MOSFET's with Very Small Physical Dimensions' (IEEE Journal of Solid-State Circuits); named after the lead author and observed to break down around 2005.
3 languages, as recorded on Wikidata. These are names the idea already goes by, not translations we made. The whole index by language.
Everything above traces to en.wikipedia.org and research.ibm.com, two sources in all, one of them the original publication rather than an account of it. Nothing is written from memory, and the method is written down. Spot an error or a better source? Suggest a fix.