Optical Frequency Division (OFD)

• Optical frequency division is the process of transferring phase stability from optical references to lower-frequency signals through precise integer division using an optical frequency comb.
• This method leverages phase-noise suppression, where fluctuations are reduced by the square of the division factor, enhancing precision in metrology and clockwork applications.
• Integrated implementations, including self-referenced, two-point, and passive Kerr approaches, enable compact, efficient, and stable ultralow-noise microwave and mmWave generation.

Optical frequency division (OFD) is the coherent transfer of the phase stability and fractional-frequency stability of an optical reference to a lower-frequency signal by exact integer division, usually through an optical frequency comb. In the literature, the term covers both optical-to-optical conversion, with $f_{\mathrm{out}} = f_{\mathrm{in}}/R$, and optical-to-microwave or millimeter-wave synthesis, in which a comb repetition rate or related electrical carrier becomes a divided representation of an optical frequency difference. The defining consequence is that phase fluctuations are suppressed by the division ratio, ideally by $N^2$ in phase-noise power spectral density, which is why OFD is central to precision metrology, optical clockwork, and ultralow-noise microwave and mmWave generation (Yao et al., 2016, Miao et al., 12 Jan 2026).

1. Core relations and operating principle

In an ideal comb, the optical mode frequencies are written as

$f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$

or equivalently

$f_n = f_0 + n f_{\mathrm{rep}},$

depending on notation. OFD uses these evenly spaced comb lines as a coherent frequency ruler: if one or two comb teeth are constrained by optical references, the repetition rate $f_{\mathrm{rep}}$ becomes the divided-down microwave or mmWave output (Miao et al., 12 Jan 2026, Liu et al., 22 Jul 2025).

A standard two-point form uses two optical references. With beat relations

$$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$

their difference gives

$\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$

so that, when locked to a local oscillator,

$f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$

This is the central OFD relation: the optical frequency difference is divided by an integer comb-index separation into the repetition rate (Sun et al., 2023).

The associated phase-noise law is the familiar division law. For Kerr OFD, one formulation is

$$\theta_{\mathrm{rep}}(t)=\frac{1}{N}\phi_\Delta(t)+\frac{\psi^\star}{N},$$

with the single-sideband phase noise reduced inside the locking bandwidth according to

$L_{\mathrm{rep}}(f)=L_\Delta(f)-20\log_{10}N.$

Equivalent statements appear throughout the literature as $N^2$0 or $N^2$1 (Egbert et al., 21 Jan 2026, Sun et al., 2023).

2. Architectures from self-referenced comb division to two-point OFD

A classical self-referenced implementation uses a comb with stabilized repetition rate and carrier-envelope offset. In a Kerr-microresonator optical clockwork, the central relation is

$N^2$2

with one comb tooth phase-locked to an optical reference and $N^2$3 measured by $N^2$4-to-$N^2$5 interferometry. A Si$N^2$6N$N^2$7 microresonator supporting a 1 THz soliton comb was used in this manner, and electro-optic phase modulation of the entire comb generated additional CW modes between the Si$N^2$8N$N^2$9 comb modes, operationally reducing the repetition frequency to electronically accessible values such as $f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$0 and $f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$1 (Drake et al., 2018).

A distinct lineage is the optical-to-optical divider. A low-noise optical frequency divider based on an optically referenced Ti:sapphire femtosecond comb, transfer-oscillator signal processing, and digital synthesis realized division instability of $f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$2 at 1 s and a fractional frequency division uncertainty of $f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$3. In that system, a cavity-stabilized 1064 nm Nd:YAG laser served as the input reference, and comparison against the fundamental/second-harmonic relationship at 1064 nm and 532 nm showed that the divider contributed uncertainty nearly three orders of magnitude below the most accurate optical clocks (Yao et al., 2016).

Two-point OFD (2P-OFD) simplifies the architecture by locking two comb endpoints to two optical references, so that the comb need not be self-referenced. This removes the need for an octave-spanning comb and an $f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$4-2$f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$5 interferometer. A microcomb-based 2P-OFD system with a high-$f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$6 all-solid-state Fabry–Pérot cavity demonstrated that one comb endpoint could be a frequency-agile single-mode dispersive wave emitted by a bright soliton microcomb. In that implementation, the dispersive wave improved beatnote SNR by 30 dB, the optical separation was about 3 THz, and the resulting microwave repetition rate was near 20 GHz (Ji et al., 2024).

3. Integrated and all-optical chip implementations

Integrated OFD separates optical referencing and nonlinear comb generation into distinct photonic subsystems. One miniaturized system combined a large-mode-volume planar-waveguide coil cavity with a waveguide-coupled Si$f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$7N$f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$8 microresonator and a high-speed CC-MUTC photodiode. The integrated reference cavity was a thin-film SiN 4-meter-long coil cavity with intrinsic $f_n = n f_{\mathrm{rep}} + f_{\mathrm{ceo}},$9 at 1550 nm and $f_n = f_0 + n f_{\mathrm{rep}},$0 at 1599 nm, while the microresonator had FSR $f_n = f_0 + n f_{\mathrm{rep}},$1 GHz and loaded $f_n = f_0 + n f_{\mathrm{rep}},$2. Optical-domain OFD was tested at $f_n = f_0 + n f_{\mathrm{rep}},$3, and at $f_n = f_0 + n f_{\mathrm{rep}},$4 the repetition-rate phase noise was about 36 dB below the optical reference phase noise below 100 kHz offset, matching the OFD prediction (Sun et al., 2023).

A fully passive Kerr version dispensed with a conventional servo loop. In a CMOS-compatible SiN platform, stable reference lasers at 1551 nm and 1600 nm were locked to the same integrated coil cavity, and OFD occurred spontaneously through Kerr interaction between the injected reference lasers and a soliton microcomb in a SiN racetrack resonator. The division relation was

$f_n = f_0 + n f_{\mathrm{rep}},$5

with the demonstrated case using $f_n = f_0 + n f_{\mathrm{rep}},$6 and a 109.5 GHz repetition rate. The measured repetition-rate shift versus injection-laser frequency followed a slope of approximately $f_n = f_0 + n f_{\mathrm{rep}},$7, and with only $f_n = f_0 + n f_{\mathrm{rep}},$8 of injected reference power the maximum locking bandwidth was estimated at about 30 MHz (Sun et al., 2024).

An earlier on-chip all-optical realization used a single continuous-wave laser to drive both a microresonator-based optical parametric oscillator and a Kerr-comb microresonator on the same chip. The beat frequency of the signal and idler fields from the OPO provided a dual-point frequency reference, and the two distinct dynamical states of Kerr cavities were passively synchronized so that broadband frequency locking transferred the stability of the OPO frequencies to the Kerr-comb repetition rate. That experiment reported a 630-fold phase-noise reduction when the Kerr comb was synchronized to the OPO, described as the lowest noise generated on the silicon-nitride platform (Zhao et al., 2023).

Active integrated actuation has also been introduced. A PZT-integrated Si$f_n = f_0 + n f_{\mathrm{rep}},$9N$f_{\mathrm{rep}}$0 racetrack resonator with FSR 109.5 GHz, intrinsic $f_{\mathrm{rep}}$1, and loaded $f_{\mathrm{rep}}$2 used an integrated high-speed PZT stress-optic actuator to tune the resonance frequency and independently adjust the soliton repetition rate without perturbing the comb offset. The measured tuning coefficient was 43.7 MHz/V, the small-signal 3-dB bandwidth was about 13 MHz, and OFD was demonstrated for an optical frequency difference of about 6 THz divided by 54 into a 109.5–110 GHz output (Liu et al., 22 Jul 2025).

4. Optical references, correlation engineering, and technical noise

The quality of the optical reference determines the eventual OFD performance. One compact two-point OFD system used co-self-injection locking (co-SIL) of two commercial DFB lasers to a single ultralow-noise miniature Fabry–Perot cavity with $f_{\mathrm{rep}}$3 and a measured ring-down lifetime of about $f_{\mathrm{rep}}$4. Because both lasers were locked to the same physical cavity, their noise was strongly common-mode correlated, and the co-SIL dual-laser reference achieved a relative phase noise of $f_{\mathrm{rep}}$5 dBc/Hz at 10 kHz offset for a 2.5 THz beatnote. Each locked laser showed over 60 dB of frequency-noise suppression relative to the free-running state, with linewidth narrowing from 16.0 kHz to 90.7 mHz Lorentzian linewidth and integral linewidth reduced to 144.5 Hz (Miao et al., 12 Jan 2026).

Another route replaces cavity-stabilized lasers with common-cavity bi-color Brillouin lasers. In that approach, the optical references operate at the fundamental quantum noise limit, and fitting the measured phase-noise spectrum yielded a Schawlow–Townes linewidth of 16.8 $f_{\mathrm{rep}}$6. Because the references were already extremely quiet, the OFD divider used an electro-optic modulation comb with the unusually small division factor $f_{\mathrm{rep}}$7, bridging an approximately 100 GHz reference gap with a 10 GHz dielectric resonator oscillator (Hu et al., 30 May 2025).

The literature is equally explicit about technical limitations. In high-precision optical division, carrier-envelope offset noise, light-path fluctuation, DDS bandwidth limitations, and comb referencing quality remain important; when $f_{\mathrm{rep}}$8 was left free-running in one optical divider, division instability degraded to about $f_{\mathrm{rep}}$9 at 1 s and $$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$0 at 10 s, and unoptimized optical paths prevented averaging below $$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$1 (Yao et al., 2016). In compact microwave OFD, servo bumps and AM-to-PM conversion recur. The co-SIL mini-FP system retained a servo bump near 200 kHz offset because the 2P-OFD stage still used an electronic phase-locked loop, and a 300 GHz Kerr OFD experiment identified AM-to-PM conversion in the UTC photodiode and RF chain as a practical limit on the observed high-offset floor (Miao et al., 12 Jan 2026, Egbert et al., 21 Jan 2026).

5. Microwave, millimeter-wave, and sub-terahertz performance

The performance range of OFD now extends from 10 GHz microwaves to 300 GHz carriers. In a simplified small-$$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$2 system referenced to quantum-noise-limited Brillouin lasers, 10 GHz microwave synthesis reached $$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$3 dBc/Hz at 1 Hz, $$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$4 dBc/Hz at 10 kHz, and $$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$5 dBc/Hz at 10 MHz offset, with RMS time jitter of 1.15 fs integrated from 10 kHz to 80 MHz. The output exhibited about 20 dB noise suppression relative to the 100 GHz reference beat note, consistent with the $$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$6 reduction expected from the OFD law (Hu et al., 30 May 2025).

At 20–50 GHz, two-point microcomb OFD has produced very strong close-in performance. A dispersive-wave-agile 2P-OFD source near 20 GHz reported, when scaled to a 10 GHz carrier, $$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$7 dBc/Hz at 100 Hz offset, $$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$8 dBc/Hz at 1 kHz offset, and $$\Delta_1 = f_A - (f_p + n f_r), \qquad \Delta_2 = f_B - (f_p + m f_r),$$9 dBc/Hz at 10 kHz offset (Ji et al., 2024). A co-SIL miniature Fabry–Perot reference combined with an integrated soliton microcomb generated a 50 GHz microwave with $\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$0 dBc/Hz at 4 kHz offset in the 50 GHz carrier domain, which scales to $\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$1 dBc/Hz at 4 kHz at 10 GHz, and $\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$2 dBc/Hz at 100 Hz offset on the 50 GHz carrier (Miao et al., 12 Jan 2026).

At 100–110 GHz, integrated Kerr systems have reported record-low results for chip-based oscillators. One integrated OFD oscillator produced 100 GHz mmWave with 9 dBm output power and phase noise of $\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$3 dBc/Hz at 10 kHz offset, corresponding to about $\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$4 dBc/Hz at 10 GHz carrier scaling (Sun et al., 2023). A Kerr-synchronized 109.5 GHz oscillator reported about $\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$5 dBc/Hz at 100 Hz offset and $\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$6 dBc/Hz at 10 kHz offset at the 109.5 GHz carrier, corresponding to approximately $\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$7 dBc/Hz at 100 Hz and $\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$8 dBc/Hz at 10 kHz when scaled to a 10 GHz carrier; the latter was identified as the lowest phase noise yet reported for an integrated photonic microwave/mmWave oscillator (Sun et al., 2024). With PZT-integrated control, a related 109.5–110 GHz system reported $\Delta = (f_A - f_B) - (N f_r), \qquad N=n-m,$9 dBc/Hz at 10 kHz offset for the 110 GHz carrier, equivalently $f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$0 dBc/Hz when scaled to 10 GHz (Liu et al., 22 Jul 2025).

At 300 GHz, OFD has entered the attosecond regime. OFD of an optically carried 3.6 THz reference down to 300 GHz through a dissipative Kerr soliton microcomb yielded a measured 300 GHz carrier phase noise of $f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$1 at 10 kHz, with timing noise about $f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$2 at 10 kHz for the generated 300 GHz carrier and fractional frequency stability improving from $f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$3 at 1 s in the free-running comb to $f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$4 at 1 s when locked to the 3.6 THz reference (Tetsumoto et al., 2020). A later Kerr optical frequency division experiment using a 3.3 THz dual-wavelength Brillouin reference injection-locked to a Kerr soliton microcomb reported a 300 GHz carrier with phase noise floor of $f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$5 dBc/Hz at 1 MHz offset and RMS timing jitter of 135 as from 1 kHz to 1 MHz, consistent with photodetection shot noise (Egbert et al., 21 Jan 2026).

6. Quantum limits, related methods, and terminological ambiguities

Recent work has extended OFD theory beyond semiclassical noise models. A comb-line-resolved quantum treatment of directly photodetected OFD derives the standard quantum limit

$f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$6

showing that the OFD quantum limit depends explicitly on the comb spectral envelope, not solely on total power. In that framework, Gaussian and sech envelopes give phase-noise suppression that improves roughly quadratically with bandwidth, $f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$7, whereas a flat-top spectrum gives only linear suppression, $f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$8. The same analysis identifies two routes below the SQL: independent intra-comb-line squeezing and inter-comb-line EPR entanglement of symmetric line pairs (Shin et al., 15 May 2026).

OFD also serves as the reference point for related methods. Electro-optic frequency division (eOFD) is described as a streamlined alternative in which an electro-optic comb is stabilized to an opto-terahertz reference rather than using a full comb-based OFD chain. A compact 300 GHz dual-wavelength Brillouin laser reference divided to 10 GHz via eOFD achieved $f_r = \frac{f_A - f_B - f_{\text{LO1}}}{N}.$9 dBc/Hz at 1 kHz, $$\theta_{\mathrm{rep}}(t)=\frac{1}{N}\phi_\Delta(t)+\frac{\psi^\star}{N},$$0 dBc/Hz at 10 kHz, and $$\theta_{\mathrm{rep}}(t)=\frac{1}{N}\phi_\Delta(t)+\frac{\psi^\star}{N},$$1 dBc/Hz at 10 MHz, while a feed-forward eOFD architecture preserved octave-spanning tunability from 8 to 16 GHz with phase noise below $$\theta_{\mathrm{rep}}(t)=\frac{1}{N}\phi_\Delta(t)+\frac{\psi^\star}{N},$$2 at kilohertz offsets and high-frequency floor between $$\theta_{\mathrm{rep}}(t)=\frac{1}{N}\phi_\Delta(t)+\frac{\psi^\star}{N},$$3 and $$\theta_{\mathrm{rep}}(t)=\frac{1}{N}\phi_\Delta(t)+\frac{\psi^\star}{N},$$4 (Egbert et al., 27 May 2025, Greenberg et al., 17 Feb 2026). These systems are not OFD in the strict comb-division sense, but they are evaluated against OFD-level spectral purity.

A separate ambiguity is purely terminological. In some photonic integrated sensing literature, “OFD” denotes “optical frequency detector,” not optical frequency division. One thin-film lithium niobate chip for real-time measurement of optical frequency variations explicitly defines OFD in that sense and reports a 5.5 mm $$\theta_{\mathrm{rep}}(t)=\frac{1}{N}\phi_\Delta(t)+\frac{\psi^\star}{N},$$5 2.7 mm optical frequency detector with 2 MHz resolution and speed up to 2500 THz/s (Yao et al., 26 Jan 2025). This usage is unrelated to frequency division and can obscure comparisons across subfields.

Taken together, the recent literature shows a field spanning self-referenced comb metrology, two-point microcomb division, passive Kerr synchronization, miniature cavity and Brillouin references, and first-principles quantum noise theory. A plausible implication is that future OFD systems will be differentiated less by the basic division law—which is now well established—than by reference engineering, passive versus servoed synchronization, quantum-limited readout, and the degree to which the full chain can be reduced to compact integrated photonics.