2026-07-12
A ring oscillator or LC tank runs at its natural frequency f₀, set by transistor delays or L·C values. Inject a small periodic signal at a frequency f_inj near f₀, and something remarkable happens: the oscillator abandons its own frequency and locks to the injected one. This is injection locking, and it's how modern SerDes, mm-wave radios, and clock multipliers achieve PLL-like behavior without the loop filter, phase detector, or settling time.
The mechanism is a phase-perturbation feedback. Each cycle, the injected current arrives slightly ahead of or behind the oscillator's own zero-crossing. That mismatch nudges the next zero-crossing earlier or later. If f_inj is close enough, the accumulated nudges exactly cancel the natural frequency drift, and the oscillator runs at f_inj with a fixed phase offset.
The range over which locking works is called the lock range, given by Adler's equation:
Where Q is the tank quality factor and I_inj/I_osc is the ratio of injected current to oscillator current. A 10 GHz LC oscillator with Q=10 and 5% injection strength locks over ±25 MHz. Push injection to 20%, and the range grows to ±100 MHz — but you're now burning more power and degrading phase noise inside the lock range.
Real-world example: Intel's mm-wave 60 GHz radios use injection-locked frequency dividers (ILFDs) to divide a 60 GHz VCO output down to a manageable 15 GHz for the PLL feedback path. A conventional CML divider at 60 GHz would burn ~30 mW; an ILFD achieves the same divide-by-4 with under 3 mW because it's just a free-running 15 GHz oscillator being whispered to by the 60 GHz source. The tradeoff: the ILFD only works over a narrow band, so you need calibration to align the free-running frequency with the target.
Rule of thumb: Injection locking phase noise inside the lock range is dominated by the injected signal, not the oscillator itself. Outside the lock range, the free-running noise takes over. This makes ILOs excellent phase-noise cleaners — an ultra-low-noise reference can "purify" a noisy oscillator over its lock range.
The dark side: if f_inj drifts outside the lock range, the oscillator doesn't just lose lock — it enters a quasi-lock beat, generating messy intermodulation products that trash spectral purity. Designers add lock detectors to disable the injection path when unlocked.