English / 日本語
← Back to the Fundamentals home Chapter 6: Part 2 · How you write the contract changes the machine

Chapter 6　Low power — how the way you write the contract bears on power

The aim of this chapter. In the preface we said, “even with the same contract, a smartphone’s low-power core and a PC-class high-performance core become entirely different machines.” One big dividing line for that is power. In this chapter we look at what a CPU spends power on, and how the designer holds power down through the machine’s build and how the contract is written. Chapter 4’s “simplicity” comes into play once more here.

6.1 Why is power so important?

Smartphones, earbuds, sensors, wristwatches — many of today’s chips run on batteries and hold heat in your palm. Power is, alongside speed, the most important constraint of design. Even if the same computation can be done, if the power is double, the battery lasts half. “How little power it runs on” decides whether that chip is usable.

6.2 What does a CPU spend power on?

Roughly, there are two sources. One is dynamic power — a little electricity spent each time a transistor switches 0↔1. The more, and the faster, it switches, the more it spends. Usually this is the lead. The other is static power — electricity that just leaks from the circuit, slowly, merely from being powered on. The larger the circuit, the more it constantly leaks a little.

Dynamic power is, roughly, decided by the multiplication of “circuit size × voltage² × speed (frequency)” — remembering it that way makes things clearer. The figure below shows where each takes effect.

The sources of power, and where reductions take effect Dynamic power is roughly decided by the multiplication of circuit size, voltage squared, and speed (frequency). Making the circuit small is where RISC works. Voltage takes effect as a square. Speed is proportional. Static power is leakage from being left on; for parts you don't use, cut power or stop the clock. Dynamic power ≈ circuit size × voltage² × speed (frequency) Circuit size small = RISC works × Voltage² squared = lowering helps a lot × Speed (frequency) lower it, drops proportionally Static power = leakage from being left on → for unused parts, cut power or stop the clock (turn off an idle stove)
Dynamic power is roughly decided by “circuit size × voltage² × speed.” Lowering any of them reduces power (voltage especially, as a square). Static power is “leakage from being left on,” held down by stopping the parts you don’t use.

6.3 Where do you move to get low power? — the designer’s levers

Looking at this multiplication and the leakage, where things take effect comes into view. The levers the designer holds are mainly four.

The designer decides here. What: at what speed, at what voltage, to run. How far to trim unneeded parts. Where, and when, to stop. How to decide: within the range that meets the speed the use demands, keep voltage and frequency as low as possible. For parts idle for long stretches, put in a mechanism to stop them. Why: because power takes effect by voltage² and speed, so running faster and at higher voltage than necessary bounces straight back as battery and heat.

6.4 Even with the same contract, a machine can be low-power or high-performance

Here, the preface’s point comes full circle. A smartphone’s low-power core and a PC-class high-performance core, even with the same instruction set (contract), have completely different inner builds. The low-power core is built slow, low-voltage, small, and stops things frequently. The high-performance core pushes with speed and many parts (spending that much power). Share the contract, choose the machine — low power is the clearest appearance of that “choosing.”

That RISC-V is favored in smartphone low-power cores and small battery-run devices is because it can be built simple and small, and suits low power. The “simplicity” and “tollgate-free freedom” seen in Chapters 4 and 5 come into effect here in the form of power.

Which decision is this knowledge for. Speed, area, power are always a tug-of-war (this is also the sequel to Chapter 2’s clock story). Once you know where low power takes effect — small, low-voltage, low-frequency, stop — you can, from “how long you want to run it on a battery,” work backward to choose the core and how to run it. This is the very core of “Physical” (clock in MHz, power, area budget) on the preface’s “decisions map.” In the next Part 3, we look concretely at “choosing the machine only as much as you need,” with RISC-V’s modular ISA.