Microring Resonators and Optical Cavities on Chip

Microring resonators are no longer a laboratory curiosity — they are a manufactured component shipping inside data-centre optics. In silicon photonics, dense banks of ring modulators and filters implement wavelength-division-multiplexed links: one fibre carries dozens of channels, each imprinted and dropped by its own micron-scale ring on the ITU grid, with companies such as Ayar Labs and Intel building optical input-output engines around this architecture. Because a ring modulator recycles the light through $\sim\mathcal{F}/\pi$ laps, it switches with far less drive energy than a straight-waveguide interferometer of the same length — what makes terabit-per-second interconnects power-viable, at the cost of a narrow, thermally twitchy window needing active stabilisation. The most spectacular modern instrument is the Kerr soliton microcomb. Pump a low-loss, anomalous-dispersion silicon-nitride ring above the parametric threshold and cascaded four-wave mixing (the $\chi^{(3)}$ physics of aq-optics-nonlinear-fiber-optics) fills the resonance ladder with hundreds of equally spaced lines; detune carefully and the field locks into a dissipative Kerr soliton — a self-organised pulse circulating indefinitely, emitting a coherent chip-scale frequency comb. Groups at EPFL, NIST, and Caltech have turned these into astrocombs calibrating exoplanet spectrographs, parallel coherent transmitters, and compact optical clocks and LiDAR engines — enabled by the dispersion engineering of aq-optics-dispersion-management, sculpting the cross-section so $n_g(\lambda)$ gives the flat, anomalous dispersion the soliton needs.

Picture the ring as a racetrack for photons. A wave nudged in from the bus circulates the loop; if after one lap its phase matches the freshly arriving wave, laps stack coherently and the internal intensity climbs far above the input — this build-up is what makes a ring a resonator rather than a bent wire. Off resonance the laps arrive out of step and the ring stays dark. The bus–ring gap is a beam-splitter you cross once per lap, and the steady state is a competition between how fast the coupler ($|\kappa|^2$ out per pass) drains the ring and how fast the round-trip loss $a=e^{-\alpha L/2}$ eats the circulating field.

When the two are matched — critical coupling — the field leaking back out of the ring is exactly equal in size and opposite in phase to the wave that ran straight down the bus, and they annihilate: every photon of input power is dissipated inside the ring and the through port goes dark. Off that balance the dip becomes shallow again, either because coupling is too weak to drain the ring (under-coupled) or too strong for the loss (over-coupled). These three regimes are the whole story of a single-bus ring, plotted next as transmission curves.

A high-$Q$ resonance is a long memory. The photon lifetime $\tau=Q/\omega_0$ reaches tens of nanoseconds in a chip-scale ring and microseconds in the best whispering-gallery cavities, so a photon covers metres to kilometres of effective path inside a device you could hide under a grain of sand. That stored path is the lever behind every application: it amplifies the index shift a bound molecule produces, multiplies the interaction length a nonlinear process needs, and turns a small tuning of $n_{\mathrm{eff}}$ into a large swing of transmission. Whispering-gallery modes are the same phenomenon with confinement handed to geometry: a wave clings to a curved dielectric rim by total internal reflection, spiralling around with an azimuthal number $m$ equal to the wavelengths fitting the circumference — the optical cousin of sound creeping around a domed cathedral's gallery.

The starting point is the resonance condition. Constructive build-up demands that the round-trip phase $\beta L$ (with $\beta=2\pi n_{\mathrm{eff}}/\lambda$) be a whole multiple of $2\pi$, which turns directly into a condition on wavelength and gives a discrete ladder of longitudinal modes indexed by the integer $m$.

where: $m$ — azimuthal (longitudinal) mode number, a positive integer; $\lambda_m$ — free-space resonance wavelength of order $m$; $n_{\mathrm{eff}}$ — effective (phase) index of the guided mode; $L$ — round-trip length, $L=2\pi R$ for a ring of radius $R$. An integer number $m$ of guided wavelengths must tile the circumference exactly, and because $n_{\mathrm{eff}}$ drifts with wavelength the ladder is not perfectly uniform. The spacing of that ladder is the free spectral range. Differentiating the resonance condition at fixed $L$, while honouring the wavelength dependence of $n_{\mathrm{eff}}$, produces a spacing governed by the group index rather than the phase index — the most common place a ring design goes wrong.

where: $\mathrm{FSR}$ — wavelength spacing between adjacent resonances; $n_g$ — group index; $\lambda$ — operating wavelength; $L$ — round-trip length. The comb is denser for a larger ring and a higher group index; using $n_{\mathrm{eff}}$ instead of $n_g$ inflates the spacing by the ratio $n_g/n_{\mathrm{eff}}$. The shape of each resonance follows from feeding the lossless coupler relation back on itself and summing the geometric series of all round trips. The through-port field emerges as a compact ratio whose modulus squared is the intensity transmission, depending only on $ta$ and the round-trip phase.

where: $T$ — through-port power transmission $|E_t/E_i|^2$; $a$ — round-trip amplitude transmission, $a=e^{-\alpha L/2}$ with $0<a\le1$; $t$ — self-coupling amplitude, real and lossless; $\phi$ — round-trip phase $\phi=\beta L$. $T$ swings from a bright off-resonance value near unity to a dark dip every time $\phi$ advances by $2\pi$, the dip depth set entirely by how close $t$ sits to $a$. Setting $\phi=2\pi m$ collapses the transmission to its on-resonance minimum and exposes the critical-coupling condition: the numerator vanishes exactly when the self-coupling amplitude equals the round-trip amplitude.

where: $T_{\min}$ — transmission at resonance ($\phi=2\pi m$); $t,a$ as above. Either weaker coupling (under-coupled) or stronger coupling (over-coupled) reopens the dip; only the exact match $t=a$ empties the through port. The sharpness of a resonance is the quality factor: the resonant frequency times the photon lifetime. Because the ring loses energy through two independent channels — intrinsic loss and out-coupling to the bus — the inverse quality factors add.

where: $Q$ — loaded quality factor; $\Delta\lambda_{\mathrm{FWHM}}$ — full-width at half-maximum; $\tau$ — photon lifetime; $Q_i,Q_c$ — intrinsic (loss) and coupling (external) quality factors; $\omega_0$ — resonant angular frequency. The faster channel dominates the linewidth, and critical coupling is the balance point $Q_c=Q_i$. Two closely related figures of merit fall out of the same Lorentzian expansion. The finesse counts resolvable channels per free spectral range (identical to a Fabry–Pérot etalon under $r^2\to ta$), and the on-resonance build-up is the factor by which the circulating intensity exceeds the input; both depend only on $ta$.

where: $\mathcal{F}$ — finesse; $B_{\mathrm{FE}}$ — on-resonance intracavity intensity enhancement; $\kappa^2=1-t^2$ — power coupling fraction. As $ta\to1$ both diverge: a low-loss, weakly-coupled ring stores light for many laps and concentrates far more intensity than sits in the bus — which is what boosts Kerr nonlinearity and sensing shifts, letting comb generation run at milliwatt pump powers. Finally, the intrinsic quality factor ties resonance sharpness back to raw waveguide loss. Since the photon lifetime is just the time for propagation loss to drain the stored energy, $Q_i$ is fixed by the attenuation coefficient, group index, and wavelength alone.

where: $\alpha$ — power attenuation coefficient of the guide; $n_g$ — group index; $\lambda_0$ — resonant wavelength. Halving the propagation loss doubles the intrinsic $Q$ — which is why the whole game of ultrahigh-$Q$ resonators is the relentless suppression of scattering and absorption.

The add-drop variant. Tap a second bus waveguide onto the opposite side, with couplers $(t_1,\kappa_1)$ and $(t_2,\kappa_2)$, and the ring becomes a channel-dropping filter. The same feedback sum applies, except a full lap now passes through both couplers so the per-lap survival factor is $t_1t_2a$. The circulating field $E_{\mathrm{ring}}=i\kappa_1E_i/(1-t_1t_2a\,e^{i\phi})$ is tapped a half-turn later by the drop coupler, giving power-complementary outputs that share one resonant denominator:

The drop port peaks — rather than dips — on resonance, with peak efficiency $T_{\mathrm{drop}}^{\max}=\kappa_1^2\kappa_2^2a/(1-t_1t_2a)^2\to1$ for a symmetric lossless ring ($a=1$, $\kappa_1=\kappa_2$): the whole selected channel routes to the drop port while the through port goes dark, the elementary WDM add-drop operation (Saleh and Teich Ch. 25).

The all-pass transmission is governed by a single competition — the self-coupling $t$ against the round-trip amplitude $a$ — and the three regimes it defines are the entire design space of a single-bus ring.

1 · Mode order and free spectral range of a silicon microring (confidence builder)

A silicon microring has radius $R=10\ \mu\mathrm{m}$, effective index $n_{\mathrm{eff}}=2.4$, and group index $n_g=4.2$ at $\lambda=1550\ \mathrm{nm}$. Find its round-trip length, the mode order near 1550 nm, and the free spectral range — then show what going wrong with the index costs you.

Step 1. The round-trip length is the circumference, $L=2\pi R=2\pi(10\times10^{-6})=62.8\ \mu\mathrm{m}$ — the path a photon retraces each lap. L = 62.8 μm.

Step 2. The mode order is the number of guided wavelengths tiling it: $m=n_{\mathrm{eff}}L/\lambda=(2.4)(62.8\times10^{-6})/(1.55\times10^{-6})=97.3$, so $m=97$, with $\lambda_{97}=n_{\mathrm{eff}}L/97=1554\ \mathrm{nm}$. m = 97, resonance ≈ 1554 nm.

Step 3. The free spectral range uses the group index: $\mathrm{FSR}=\lambda^2/(n_gL)=(1.55\times10^{-6})^2/[(4.2)(62.8\times10^{-6})]=2.40\times10^{-12}/2.64\times10^{-4}=9.1\times10^{-9}\ \mathrm{m}$. FSR = 9.1 nm.

Step 4. The trap: plugging $n_{\mathrm{eff}}=2.4$ into the same formula gives $\lambda^2/(n_{\mathrm{eff}}L)=15.9\ \mathrm{nm}$, too large by $n_g/n_{\mathrm{eff}}=1.75$ — the difference between hitting and missing the channel grid. Correct FSR 9.1 nm; using n_eff overestimates by 1.75×.

2 · Photon lifetime, linewidth, and effective path of a microtoroid

An ultrahigh-$Q$ silica microtoroid has $Q=1\times10^8$ at $\lambda_0=1550\ \mathrm{nm}$, with group index $n_g\approx1.45$. Find the photon lifetime, the resonance linewidth in both wavelength and frequency, and the effective distance a photon travels before it decays.

Step 1. The resonant angular frequency is $\omega_0=2\pi c/\lambda_0=2\pi(2.998\times10^8)/(1.55\times10^{-6})=1.22\times10^{15}\ \mathrm{rad/s}$. ω₀ = 1.22×10¹⁵ rad/s.

Step 2. The photon lifetime is $\tau=Q/\omega_0=10^8/1.22\times10^{15}=8.2\times10^{-8}\ \mathrm{s}$ — the cavity holds a photon for τ = 82 ns, an eternity by chip-scale standards.

Step 3. The linewidth is $\Delta\lambda=\lambda_0/Q=1550\ \mathrm{nm}/10^8=15.5\ \mathrm{fm}$, equivalently $\Delta\nu=\nu_0/Q=(1.93\times10^{14})/10^8=1.9\ \mathrm{MHz}$ — a resolving power of one part in $10^8$. Δλ = 15.5 fm, Δν = 1.9 MHz.

Step 4. The effective path in one lifetime is $\ell=v_g\tau=(c/n_g)\tau=(2.998\times10^8/1.45)(8.2\times10^{-8})=17\ \mathrm{m}$ — roughly $5\times10^4$ laps of a sub-millimetre rim folded into a device you can barely see. ℓ ≈ 17 m.

3 · Critical coupling, finesse, and build-up of a silicon-nitride ring

A silicon-nitride ring has propagation loss $\alpha=0.1\ \mathrm{dB/cm}$, radius $R=100\ \mu\mathrm{m}$, and $n_g=2.0$ at $\lambda_0=1550\ \mathrm{nm}$. Find the intrinsic $Q$ and, at critical coupling, the finesse and build-up.

Step 1. Convert loss to base-e: $\alpha=0.1\ \mathrm{dB/cm}=10\ \mathrm{dB/m}$, and since dB attenuation is $4.343\,\alpha L$, $\alpha=10/4.343=2.30\ \mathrm{m}^{-1}$. Then $Q_i=2\pi n_g/(\alpha\lambda_0)=2\pi(2.0)/[(2.30)(1.55\times10^{-6})]=3.5\times10^6$, and with $L=2\pi R=628\ \mu\mathrm{m}$ the round-trip amplitude is $a=e^{-\alpha L/2}=e^{-7.23\times10^{-4}}=0.99928$. Q_i = 3.5×10⁶, a = 0.99928.

Step 2. Critical coupling sets $t=a$, so $\kappa^2=1-t^2=1.43\times10^{-3}$ and $ta=a^2=0.99857$. The finesse is $\mathcal{F}=\pi\sqrt{ta}/(1-ta)=3.139/1.43\times10^{-3}\approx2.2\times10^3$, and the build-up $B_{\mathrm{FE}}=\kappa^2/(1-ta)^2\approx7.0\times10^2$. F ≈ 2200, build-up ≈ 700×.

Step 3. So a milliwatt of pump becomes hundreds of milliwatts inside the loop — enough to seed a Kerr comb. As a check, the loaded $Q=Q_i/2=1.8\times10^6$ matches $\mathcal{F}\times n_gL/\lambda_0=2200\times810$. Loaded Q = 1.8×10⁶.

The whole field of microresonators traces to a piece of Victorian acoustics — a puzzle about why a whisper could carry around the gallery of a cathedral dome — that turned out to describe how light clings to the rim of a glass sphere.

Texts. Saleh and Teich, Fundamentals of Photonics, 3rd ed., Ch. 25 (Resonators) and Ch. 24 (Photonic Devices) · Hecht, Optics, 5th ed., §9.6 (Fabry–Pérot interferometer, finesse and FSR) · Loudon, The Quantum Theory of Light, 3rd ed., Ch. 1 (cavity modes and photon lifetime) · Foot, Atomic Physics, Ch. 7 (optical cavities and the Q factor) · Demtröder, Laser Spectroscopy, Vol. 1, Ch. 4 (interferometers, finesse, spectral resolution) · Bogaerts et al., "Silicon microring resonators," Laser and Photonics Reviews 6, 47 (2012).

Provenance. The all-pass and add-drop transfer functions and the $Q$/finesse relations follow the standard treatment (Bogaerts et al. 2012; Saleh and Teich Ch. 25). A verifier should reproduce both from the lossless coupler matrix with $t^2+\kappa^2=1$, confirm the Fabry–Pérot correspondence $ta\to r^2$ (Hecht §9.6), and recompute the worked numbers: $\mathrm{FSR}=9.1\ \mathrm{nm}$ ($R=10\ \mu\mathrm{m}$, $n_g=4.2$); $\tau=82\ \mathrm{ns}$, $\Delta\nu=1.9\ \mathrm{MHz}$ ($Q=10^8$); $Q_i=3.5\times10^6$, $\mathcal{F}\approx2200$, build-up $\approx700$ ($\alpha=0.1\ \mathrm{dB/cm}$). The $Q$-partition ($1/Q=1/Q_i+1/Q_c$, critical coupling $\Leftrightarrow Q_c=Q_i$) follows from the damped-oscillator model (Loudon Ch. 1; Foot Ch. 7). All checks recompute from the stated inputs with no non-standard constants.

Prerequisites. Waveguide Modes and the Wave Equation · Coupled-Mode Theory and Directional Couplers · Total Internal Reflection and Optical Guiding · Fiber Attenuation and Loss Mechanisms. Leads to. Integrated Photonic Waveguides and Circuits · Nonlinear Optical Fibers and Effects · Dispersion Management and Compensation.

Ask the tutor: ask me to walk through why the same $t=a$ condition that zeroes the through port also means the coupling and loss rates are equal, and I'll build it from both the transfer function and the photon-lifetime picture.