Chip-Scale Atomic Clock (CSAC)
Atomic Timing in a Matchbook-Sized Package
The CSAC emerged from a DARPA program in the early 2000s with a deceptively simple goal: deliver atomic-clock stability while consuming the power of a single LED. Conventional atomic standards, whether cesium-beam or rubidium, rely on a physical microwave resonant cavity and a discharge lamp, which fix a floor of several watts and tens of cubic centimeters. The CSAC eliminates the cavity entirely through an all-optical interrogation scheme, allowing the physics package to shrink to a few cubic millimeters and the complete module to fit in roughly 17 cm³.
At the heart of the device, a vertical-cavity surface-emitting laser (VCSEL) illuminates a micro-fabricated cell containing cesium-133 vapor and a buffer gas. The laser is current-modulated at 4.596 GHz, exactly half the cesium ground-state hyperfine splitting, so its two first-order optical sidebands are separated by the full 9.192631770 GHz transition. When that separation is precise, the atoms are pumped into a coherent dark state and stop absorbing light. A photodiode behind the cell senses the resulting transmission peak, and a control loop steers the local oscillator to hold the resonance, transferring atomic accuracy to the 10 MHz electrical output.
Because the entire scheme is optical and the heated mass is tiny, the CSAC reaches operating temperature in two to three minutes and idles below 150 mW. Its weakness is short-term phase noise and vibration sensitivity, since the small cell and low signal-to-noise ratio limit the loop bandwidth. Designers therefore often pair a CSAC with a low-noise quartz oscillator, letting the atomic reference correct long-term drift while the quartz handles close-in noise.
Governing Physics and Performance Equations
fCs = 9,192,631,770 Hz (defines the SI second)
VCSEL modulation frequency (CPT):
fmod = fCs / 2 ≈ 4.596 GHz (sideband half-splitting)
Fractional frequency to time error:
Δt ≈ σy(τ) × τ (holdover drift over interval τ)
Example: with σy ≈ 1 × 10−11 averaged near τ = 104 s, accumulated error stays on the order of 1 μs over several hours of GPS-denied holdover.
CSAC Versus Other Frequency References
| Reference | Volume | Power | ADEV @ 1 s | Warm-up | Holdover to 1 μs |
|---|---|---|---|---|---|
| CSAC (cesium CPT) | ≈ 17 cm³ | < 150 mW | 3 × 10−10 | 2 to 3 min | 2 to 4 hours |
| Rubidium standard | 50 to 100 cm³ | 5 to 15 W | 1 × 10−11 | 3 to 5 min | 4 to 10 hours |
| OCXO | 10 to 50 cm³ | 1 to 3 W | 1 × 10−12 | 3 to 10 min | minutes to ~1 hour |
| TCXO | < 1 cm³ | 10 to 50 mW | 1 × 10−9 | seconds | seconds |
Frequently Asked Questions
How does a CSAC achieve atomic stability with under 150 mW of power?
It drops the bulky microwave cavity of a cesium-beam clock and uses coherent population trapping instead. A VCSEL on the cesium D1 line at 894.6 nm is modulated at 4.596 GHz, so its two optical sidebands span the 9.192 GHz hyperfine transition and drive a dark resonance in a millimeter-scale vapor cell. With no resonant cavity and only a tiny heated mass, the whole module draws about 120 to 150 mW versus several watts for rubidium and tens of watts for a lab cesium-beam clock.
What Allan deviation and holdover can a CSAC deliver during GPS outage?
A unit like the Microsemi SA.45s specifies an Allan deviation near 3 × 10−10 at 1 s, dropping to roughly 1 × 10−11 at 1000 s, with aging below 9 × 10−10 per month. From a GPS-disciplined start that yields about 1 μs of time error after 2 to 4 hours of holdover and tens of microseconds per day. See Allan deviation for how that metric is measured.
How does a CSAC compare to an OCXO or rubidium oscillator for a portable design?
An OCXO has excellent short-term stability and low phase noise but drifts into the microsecond range over hours. A rubidium standard edges out the CSAC on long-term stability but needs 5 to 15 W and a larger case. The CSAC wins the power-volume-stability product: about 17 cm³, under 150 mW, a 2 to 3 minute warm-up, and 1 μs per several hours of holdover, ideal where an OCXO is not stable enough and rubidium is too power-hungry.