Dennard scaling

Dennard scaling (Robert Dennard, 1974) is the principle that as transistors shrink, their power density remains constant: voltage and current scale down proportionally with linear dimensions, so the power per unit area stays fixed even as the number of transistors per unit area doubles. Dennard scaling was the companion law to Moore's Law that made exponential transistor growth practical. Without it, doubling transistors every two years would have meant doubling power consumption every two years — a trajectory that would have melted chips within a decade.

The law held until roughly 2005-2007, when leakage currents and threshold voltage constraints made further voltage scaling impossible. Clock speeds stagnated around 3-4 GHz, and the industry pivoted to multicore architectures to continue extracting performance without increasing single-core frequency. The death of Dennard scaling was more immediately consequential than the slowing of Moore's Law itself: it ended the free lunch of single-threaded performance growth and forced a fundamental rearchitecture of software around parallelism.

Dennard scaling's collapse is the reason that modern performance gains come from specialization rather than general-purpose speedup. The era of 'faster cores' ended; the era of 'more cores, specialized cores, and heterogeneous computing' began. It is the unsung inflection point that shaped contemporary computer architecture more than any other single event.