end-of-dennard-scaling

The end of Dennard scaling refers to the loss of a historically useful constant-field scaling pattern. Dennard and colleagues’ 1974 MOSFET paper described relationships for reducing device dimensions together with voltage. Under the idealized relationships, transistor density and switching speed could improve while power density remained approximately controlled. The model relies on coordinated changes, not just smaller lateral dimensions.

Supply and threshold voltages could not keep falling in the same proportion indefinitely. Lower threshold voltage increases subthreshold current, while lower supply voltage reduces available overdrive and noise margins. Device electrostatics and thin-dielectric tunneling introduce further constraints. As voltage reduction slowed relative to density and frequency growth, the old power assumptions became less favorable.

End of Dennard scaling Conceptual comparison; not a manufacturing specification or scale drawing Ideal scaling Dimensions and voltage shrink Approximately controlled power density Constraint Voltage scaling slows Leakage and margins limit options Design response Parallelism + specialization Control activity and operating point The breakdown concerns a coordinated voltage-and-dimension scaling pattern, while other kinds of semiconductor progress continue.

Switching-power arithmetic shows the consequence. A common first-order model is dynamic power proportional to activity × capacitance × voltage squared × frequency. Increasing the number of active circuits and their switching rate raises total power unless other factors compensate. The equation is an approximation, but it makes the importance of supply-voltage scaling explicit.

Power limits changed architectural priorities without ending transistor development. Multiple cores, specialized accelerators, clock gating, power gating, and dynamic voltage-frequency control offer different ways to manage useful work per energy. Each has workload and implementation limits. Additional transistors can still be valuable even when they cannot all operate at maximum activity simultaneously.

The transition was gradual and technology dependent. It is more accurate to identify the failed assumptions and resulting design trade-offs than to assign every product one universal cutoff year. Gate-stack innovations and new device structures continued to improve particular metrics. The end of the simple scaling pattern made energy and thermal constraints central to process and architecture decisions.

Sources and further reading. [Dennard et al., MOSFET scaling paper, 1974](https://doi.org/10.1109/JSSC.1974.1050511); [National Research Council, The Future of Computing Performance, 2011](https://doi.org/10.17226/12980).

Read End of Dennard scaling as the loss of automatic power-density benefits from proportional device and voltage scaling.

Go deeper with CFSGPT

Get AI-powered deep-dives, save terms, and run advanced simulations — free account.

Create Free Account