Chapter 11Mach–Zehnder Interferometers

Split, delay, recombine: the Mach–Zehnder interferometer turns a phase difference into a change of intensity, and it is everywhere in photonics.

In this chapter you will learn to
  • derive the transfer function of balanced and unbalanced MZIs from coupler and waveguide matrices;
  • relate the free spectral range to the group index and the arm-length difference;
  • find the extinction ratio for non-ideal splitters and unequal losses;
  • describe tunable MZI switches and lattice filters;
  • use an MZI as a test structure to measure the group index.
Design question

How does the arm-length difference set the spectrum, and what limits the extinction?

11.1Interference and optical phase

A wave that travels a length \(L\) of waveguide picks up a phase \(\phi = \beta L = 2\pi n_\mathrm{eff}L/\lambda\). Two waves that meet add as fields, so the result depends on their phase difference: in phase they interfere constructively, a half-wave apart destructively. An interferometer splits light into two paths, lets their phases differ, and recombines them, turning a phase difference into a change of power. Because \(\lambda\) is about 1.5 µm and \(n_\mathrm{eff}\) about 2.4, a length difference of only 0.32 µm in a silicon waveguide is already half a wavelength.

11.2The MZI transfer function

The Mach–Zehnder interferometer (MZI) is two couplers joined by two arms. Each element is a matrix acting on the pair of field amplitudes in the two waveguides (Chapter 9): a lossless coupler with power coupling \(\kappa^2\) and through coupling \(t^2 = 1 - \kappa^2\) is

\[C = \begin{pmatrix} t & -j\kappa \\ -j\kappa & t\end{pmatrix},\]

and the arms multiply each field by \(a_i e^{-j\phi_i}\), with \(a_i\) the field transmission (loss) and \(\phi_i\) the phase. The output is \(C_2\,\mathrm{diag}(a_1e^{-j\phi_1}, a_2e^{-j\phi_2})\,C_1\) applied to the input. With ideal 50/50 couplers and equal losses \(a\), light entering one port leaves the two outputs as

MZI with ideal 50/50 couplers
\[P_\mathrm{bar} = a^2\sin^2\frac{\Delta\phi}{2}, \qquad P_\mathrm{cross} = a^2\cos^2\frac{\Delta\phi}{2}, \qquad \Delta\phi = \phi_2 - \phi_1\]

The two outputs are complementary: power that leaves one appears at the other. (Which output is called “bar” and which “cross” depends on the coupler phases; with the −j convention above, a balanced MZI sends everything to the cross port.)

When the couplers are not 50/50, the two arms carry unequal fields and can no longer cancel completely, so the nulls fill in. For couplers with power ratios \(\kappa_1^2\) and \(\kappa_2^2\) the extinction ratio of the cross port is

\[\mathrm{ER}_\mathrm{cross} = \left(\frac{t_1\kappa_2 + \kappa_1 t_2}{t_1\kappa_2 - \kappa_1 t_2}\right)^2,\]

which is infinite only when the couplers are equal, \(\kappa_1^2 = \kappa_2^2\); the bar port’s nulls are perfect instead when \(\kappa_1^2 + \kappa_2^2 = 1\). Unequal arm losses unbalance the fields in the same way, which is why a long unbalanced arm slowly fills the nulls.

Worked Example 11.1 · How good must the splitters be?

With a perfect first coupler (50/50) and a second at 55/45, the cross port has \(t_1\kappa_2 = \sqrt{0.5}\sqrt{0.55} = 0.524\) and \(\kappa_1t_2 = \sqrt{0.5}\sqrt{0.45} = 0.474\), so ER = \((0.998/0.050)^2 = 398\), or 26 dB. At 60/40, ER drops to 20 dB. A switch that must block a channel by 30 dB needs couplers matched to within a few per cent, or a calibration, which is what tunable couplers and the “double MZI” are for.

11.3Unbalanced MZIs

If the two arms differ in length by \(\Delta L\), the phase difference depends on wavelength, \(\Delta\phi = 2\pi n_\mathrm{eff}\Delta L/\lambda\), and the outputs oscillate across the spectrum. The spacing between adjacent maxima, the free spectral range, follows from the change of \(\Delta\phi\) by \(2\pi\), and because \(n_\mathrm{eff}\) itself depends on wavelength it involves the group index (Chapter 2):

Free spectral range of an unbalanced MZI
\[\mathrm{FSR} = \frac{\lambda^2}{n_g\,\Delta L}\]
Worked Example 11.2 · A 100 µm imbalance

In 500 × 220 nm silicon strips at 1550 nm (\(n_g = 4.18\), Chapter 6), \(\Delta L = 100\) µm gives FSR = \(1.55^2/(4.18 \times 100) = 5.75\) nm. To separate channels 3.2 nm apart (the 400 GHz grid of the reference link) as an interleaver, you need FSR = 6.4 nm, so \(\Delta L = 2.4025/(4.18 \times 0.0064) = 90\) µm.

Try it. Set the arm-length difference, the splitter ratios, the loss and the phase; the top view shows the delay loop and where the light goes.

// try it · Mach–Zehnder interferometer (2 × 2, Si strip waveguides)
geometry: top view (glow ∝ power at 1550 nm)
transmission against wavelength
FSR –extinction (bar port) –ng 4.18excess loss –

11.4Tuning and switching

A phase shifter on one arm slides the fringes without changing their spacing. With a balanced MZI (\(\Delta L = 0\)), a phase shift of π switches the light from one output to the other: the MZI switch, the building block of switch fabrics and of the interferometer meshes used for photonic computing (Chapter 34). The phase shifter can be a heater (slow, efficient, Chapter 18) or an electro-optic section (fast, Chapters 19–20); driven with data, the same MZI is a modulator.

11.5Cascaded and lattice filters

One unbalanced MZI has a sinusoidal response, which is a poor filter: its passband is rounded and its rejection grows slowly away from the nulls. Cascading MZIs, with couplers of chosen ratios and delays that are multiples of a unit \(\Delta L\), synthesises higher-order responses, exactly as digital filters are built from delays and weights: flat-top passbands, sharp edges and interleavers that split a WDM comb into odd and even channels (Chapter 14). Each extra stage adds a coupler and a phase that must be accurate, so lattice filters are usually tuned with heaters after fabrication.

11.6MZIs as test structures

Because the FSR depends only on \(n_g\) and the drawn \(\Delta L\), an unbalanced MZI is the standard way to measure the group index on a chip: find the spacing of the fringes, and

\[n_g = \frac{\lambda^2}{\mathrm{FSR}\,\Delta L}.\]

A set of MZIs with different \(\Delta L\) gives several estimates and an error bar; the depth of the fringes measures how well the couplers are balanced. This is one of the first things you put on a test chip and one of the first analyses of Chapter 32.

Your first chip

Include two or three unbalanced MZIs between grating couplers, with \(\Delta L\) between about 50 and 300 µm (FSR from about 11 to 2 nm at 1550 nm), built from the PDK’s Y-branches or broadband couplers. They give the group index and a check on the splitters.

Explore further

Free simulations by others that pair well with this chapter [11]:

Reading

Key takeaways
  • An MZI is coupler × arms × coupler; with 50/50 couplers the outputs go as \(\sin^2(\Delta\phi/2)\) and \(\cos^2(\Delta\phi/2)\).
  • The free spectral range is \(\lambda^2/(n_g\Delta L)\), with the group index: 5.75 nm for 100 µm in a silicon strip at 1550 nm.
  • Unequal couplers or arm losses limit the extinction: a 55/45 coupler caps it near 26 dB.
  • A phase shifter turns a balanced MZI into a switch or modulator; cascades make flat-top filters and interleavers.
  • Unbalanced MZIs measure the group index on a chip: \(n_g = \lambda^2/(\mathrm{FSR}\,\Delta L)\).