Atomic clocks throw down the gauntlet to VCSEL makers

In order to achieve atomic-timing accuracy in portable applications that can run for 24 hours on a small battery, the US Defense Advanced Research Projects Agency (DARPA) launched the chip-scale atomic clock (CSAC) program in 2002. Its five-year goal is to produce a 1 cm3 atomic clock consuming 30 mW or less.

With the CSAC project now in its fifth year and several competing teams (NIST, Symmetricom/Draper/Sandia, Rockwell/Agilent, Honeywell and Sarnoff/Princeton) succeeding well beyond initial expectations, attention is turning to commercialization of the technology and potential applications. Ultimately, widespread deployment of CSAC technology will depend on the final product performance specifications and cost, but here we will consider a few application possibilities.

Potential CSAC applications include ad-hoc mobile communication networks that could self-assemble among a group of people, each carrying their own atomic clock for precise synchronization. CSAC-enhanced GPS receivers could lock to satellites much more rapidly, allowing for faster acquisition of precise positioning information. Finally, if CSACs could be mass produced using batch fabrication techniques, the unit cost could potentially fall to the point where CSACs could replace ovenized crystal oscillators in a large number of precision timing applications.

DARPA s mid-term three-year power goal of 200 mW, while less aggressive than the ultimate 30 mW goal, still represents a fifty-fold reduction in power consumption relative to current commercial atomic clocks. The team comprised of Symmetricom, the Charles Stark Draper Laboratory and Sandia National Laboratories has recently reported (see Lutwak et al. 2007 in Further reading) a pre-production build of 10 miniature atomic clocks that meet the mid-term power goal, consuming only 125 mW each. This pre-production build represents a first step toward commercialization, allowing us to determine the manufacturability, unit-to-unit variation and long-term aging effects of this new generation of atomic clocks.

Achieving the ambitious CSAC objectives of reducing the size and power of atomic clocks by two orders of magnitude demands innovation, both in the fundamental approach to atomic interrogation (see box "Switching from lamps to VCSELs") as well as in the materials and fabrication processes of the physics package assembly. Replacing the conventional RF discharge lamp with a VCSEL cuts the power required to drive the light source from several watts to several milliwatts. Incorporating the VCSEL and a photodiode with a MEMS-fabricated vapor cell permits a reduction in the size and mass of the resonance cell and consequently the necessary heater power. Figure 1d shows a MEMS physics package, developed by our colleagues at the Charles Stark Draper Laboratory, that requires only 10 mW to heat a tiny cesium vapor cell to 85 °C.

While functional CSACs have been demonstrated in the laboratory, commercialization of the technology requires development of production processes for several key components, including high-performance VCSELs and MEMS-fabricated physics packages.

The performance of the VCSEL component directly influences the performance, power consumption and reliability of the CSAC. Unfortunately, the VCSELs used today for data communication are inadequate for use in CSACs that are based on coherent population trapping. VCSELs for these clocks must emit in a single polarization state, operate in a single transverse mode, be tuned exactly to a particular resonance wavelength of the atomic vapor (typically rubidium or cesium) and remain stable at this wavelength and polarization for the life of the CSAC. They must have a low threshold current and high modulation bandwidth to reduce power consumption. Finally, because the noise properties of the VCSEL directly impact the stability of the atomic clock, they must exhibit unusually high spectral purity (narrow linewidth).

Table 1 shows the available optical resonance wavelengths, ranging from 780 to 895 nm, and microwave clock frequencies for rubidium and cesium atomic clocks. D1 wavelengths yield higher signal-to-noise ratios and are preferable to D2 wavelengths.

VCSELs can reach rubidium wavelengths using AlGaAs quantum wells, and cesium wavelengths using InGaAs quantum wells. However, these lasers must also operate at a single frequency that is tunable to exactly the D1 or D2 wavelength at a specified operating temperature. For example, if the VCSEL operating temperature is constrained to 85 °C ± 5 °C, then the wavelength is constrained to ±0.3 nm. This immediately reduces yield to at most 20% for an epitaxial growth uniformity of 3 nm (0.35%) across a 75 mm diameter GaAs wafer.

Table 2 lists most of the important VCSEL requirements for use in CSACs. To minimize DC power consumption and overhead heat dissipation, a threshold current below 0.5 mA is desired (see figure 4a). The VCSEL, which is typically operated at 0.5 mA above this threshold, should provide an optical power in the cesium cell between 20 and 40 μW, to maximize the clock signal-to-noise ratio and minimize light-induced frequency shifts. Importantly, the VCSEL linewidth should be at most 75 MHz, so that it is much narrower than the pressure-broadened cesium absorption linewidths (see figure 4b). These requirements in threshold current and linewidth can be met simultaneously by increasing the top distributed Bragg reflector reflectivity from 99.6% for a typical data communication VCSEL, to 99.9% for a CSAC VCSEL. Modulation requirements

The VCSELs used in CSACs must also be frequency modulated at half of the clock frequency using enough RF power to put most of the optical power into the two sidebands at ±4.6 GHz (modulation index M = 1.8). Ideally, the required RF power to produce sufficient modulation is less than –10 dBm to minimize power consumption.

The VCSEL also needs to operate in a single transverse and linear polarization mode, because other modes are detuned from resonance and will simply add partition noise. Finally, the VCSEL should operate for a lifetime of 10 years (90,000 hours) at an operating temperature of approximately 85 °C.

It is a significant challenge to manufacture VCSELs that meet all of these requirements simultaneously with a yield of 10% or more. Fortunately, in the near-term, atomic-clock customers can probably afford to pay ten times more for their VCSELs than the going rate for standard data communication VCSELs that dominate the market today.