Delays, timers and PWM

The blink example waits with a delay function. The simplest version of a delay is a loop that counts to a large number and does nothing else. It works in a debug build, but with optimization enabled it takes no time at all: the compiler sees a loop with no effect and removes it. Declaring the counter volatile prevents this, but the main point is that timing should come from hardware, not from counting instructions.

A hardware timer is a counter driven by a clock. Its rate is set by 2 values. Taking an STM32 as the example, the prescaler divides the input clock: with a 72 MHz clock and a prescaler value of 71, the counter ticks at 72 MHz / 72 = 1 MHz, one million times per second. The auto-reload value, called ARR, is the point at which the counter wraps to zero. The period is ARR + 1 ticks, so 1 MHz ticks with ARR = 999 give an update 1000 times a second. Other chips divide differently: AVR has a fixed set of dividers and the RP2040 has a fractional divider. Check the timer chapter of your chip's manual for the exact formula.

Nearly every Cortex-M chip also has SysTick, a simple 24-bit timer that counts down and reloads. Most libraries use it for a millisecond tick, which gives you a millis() style clock. With this clock you can wait without blocking: record when something started, and check on each pass of the loop whether enough time has passed.

PWM (pulse width modulation) is generated by a timer: the output pin goes high at the start of each period and low when the counter reaches a compare value. The fraction of the period during which the pin is high is the duty cycle. A compare value of 500 out of 2000 counts is 25 percent. When the switching is fast enough, an LED appears dimmed, a motor runs slower and a servo moves to a position. The pin is still only ever fully on or fully off.