Photonics Engineer Interview Questions, with Worked Answers

Scope

This article collects technical questions of the kind asked in interviews for photonics, optical, laser and fiber-optic engineering roles, from new graduates to experienced engineers, and gives a short answer to each with the numbers worth knowing. The questions are grouped by subject; each answer links to the site's page that treats the subject in full. Company-specific processes, behavioral questions and salaries are outside its scope.

Technical interviews in this field usually test three things: whether the candidate can do quick order-of-magnitude calculations (photon energy, dB arithmetic, beam sizes, noise), whether they understand why devices behave as they do, and whether they have worked at a bench. The answers below are written to show the reasoning an interviewer listens for, not only the result.

Fundamentals

What is the energy of a 1550 nm photon, and how many photons per second are in 1 mW? $E = hc/\lambda$, and with $hc = 1239.8$ eV·nm, $E = 1239.8/1550 = 0.800$ eV. One milliwatt is $10^{-3}$ J/s, so the photon rate is $10^{-3}/(0.800 \times 1.602\times10^{-19}) = 7.8\times10^{15}$ photons per second. Remembering 1239.8 eV·nm makes this a mental calculation. See photon energy and the wavelength calculator.

Convert 10 mW to dBm. What do two 0 dBm sources give together? $P_\text{dBm} = 10\log_{10}(P/1\,\text{mW})$, so 10 mW is 10 dBm and −20 dBm is 10 µW. Powers add linearly, not in dB: two incoherent 1 mW sources give 2 mW, which is $10\log_{10}2 = 3.01$ dBm, not 0 dBm + 0 dBm. A loss of 3 dB halves the power; 10 dB is a factor of ten. See decibel.

What reflects from an uncoated glass surface, and from silicon? At normal incidence $R = [(n_1-n_2)/(n_1+n_2)]^2$. For glass of index 1.5 in air, $R = (0.5/2.5)^2 = 4.0\%$ per surface. For silicon at 1550 nm ($n = 3.476$), $R = 30.6\%$, which is why silicon facets and windows need anti-reflection coatings. The angle and polarization dependence is in the Fresnel equations.

What does numerical aperture tell you? $\text{NA} = n \sin\theta$, the sine of the largest half-angle of light accepted or emitted, multiplied by the index of the medium. It sets a lens's resolution and spot size and a fiber's acceptance angle and, with the core size, the number of guided modes. See numerical aperture.

Optical fiber

Why do telecom systems use 1310 nm and 1550 nm? Standard single-mode silica fiber has its zero-dispersion wavelength near 1310 nm and its lowest attenuation near 1550 nm, about 0.2 dB/km. The C band around 1550 nm is also where erbium-doped fiber amplifiers provide gain, which made it the band for long-haul wavelength-division multiplexing. See fiber attenuation.

When is a step-index fiber single-mode? When the V number $V = (2\pi a/\lambda)\,\text{NA}$ is below 2.405, where $a$ is the core radius. A fiber with $a = 4$ µm and NA 0.12 has $V = 1.95$ at 1550 nm and a cutoff wavelength of $2\pi a\,\text{NA}/2.405 = 1.25$ µm; it is single-mode above about 1250 nm. The fiber NA and V number calculator does the arithmetic.

How much does a 0.1 nm wide source spread after 80 km of standard fiber? With a chromatic dispersion of 17 ps/(nm·km) near 1550 nm, the spread is $D \cdot L \cdot \Delta\lambda = 17 \times 80 \times 0.1 = 136$ ps, more than a bit period at 10 Gb/s (100 ps). The answer should note that for a modulated narrow-linewidth laser the spectral width is set by the modulation, not the laser. The dispersion tool computes the spread for other lengths and widths.

Estimate the received power over a 10 km link at 1310 nm. A link budget adds the losses: 10 km at 0.35 dB/km is 3.5 dB, two connectors at 0.5 dB each add 1.0 dB, and two splices at 0.1 dB add 0.2 dB, 4.7 dB in total. A 0 dBm transmitter then delivers −4.7 dBm, and the margin is the difference from the receiver's sensitivity. The link budget tool works through the same arithmetic for datacenter links.

Lasers and beams

What conditions does a laser need to lase? A gain medium with population inversion, and a resonator whose round-trip gain equals its round-trip loss, including the output coupling. Threshold is the pump level at which gain first reaches loss; above it, the gain clamps and additional pump power goes into output. For a diode laser this appears as the threshold current on the L-I curve; see threshold extraction methods.

A beam has a 1 mm waist radius at 1064 nm. How far does it stay collimated, and what is its divergence? For a Gaussian beam, the Rayleigh range is $z_R = \pi w_0^2/\lambda = 2.95$ m and the far-field half-angle divergence is $\theta = \lambda/(\pi w_0) = 0.34$ mrad. Halving the waist quarters the Rayleigh range and doubles the divergence. Try it in the Gaussian beam tool.

What is M², and how is it measured? The beam quality factor M² is the ratio of a beam's beam parameter product to that of an ideal Gaussian beam at the same wavelength; a perfect TEM₀₀ beam has M² = 1. It is measured by recording the second-moment beam diameter at several positions through a focus and fitting the hyperbolic caustic, as in M² beam quality measurement.

Detectors and noise

What is the highest responsivity a photodiode can have at 1550 nm? $R = \eta q \lambda/(hc)$, which is 1.25 A/W at 1550 nm for a quantum efficiency $\eta$ of 1. A typical InGaAs photodiode at 1.0 A/W therefore has $\eta = 0.80$. See responsivity and quantum efficiency.

Compare shot noise and thermal noise for 1 mA of photocurrent in 1 GHz. Shot noise is $\sqrt{2qIB} = 0.57$ µA rms. The thermal noise of a 50 Ω load at 300 K in the same bandwidth is $\sqrt{4kTB/R} = 0.58$ µA rms. The two are comparable at this current; below it, thermal noise dominates a 50 Ω receiver, which is why sensitive receivers use a transimpedance amplifier or an avalanche photodiode. The photodiode noise tool shows the crossover.

When would you choose an APD over a PIN photodiode? When the receiver is limited by amplifier noise rather than shot noise, at low optical power and high bandwidth. The APD's gain lifts the signal above the amplifier noise, but its excess noise grows with gain, so there is an optimum gain, often 10 to 15 for an InGaAs APD in a high-speed receiver. When the signal is strong enough to be shot-noise limited, a PIN is better. The choice is worked through in the avalanche photodiode entry and types of photodetectors.

Integrated photonics

What sets the free spectral range of a ring resonator? $\text{FSR} = \lambda^2/(n_g L)$, with the group index, not the effective index, because adjacent resonances differ in wavelength. A silicon ring of 10 µm radius ($L = 62.8$ µm) with $n_g = 4.3$ has an FSR of 8.9 nm at 1550 nm. See ring resonator and the ring resonator tool.

How is a Mach-Zehnder modulator biased for intensity modulation, and what is Vπ? Vπ is the voltage that changes the phase difference between the arms by π, taking the output from maximum to minimum. For linear intensity modulation the Mach-Zehnder modulator is biased at quadrature, the half-power point where the transfer function is steepest and most linear. Measuring it is covered in modulator Vπ measurement.

How do you measure the propagation loss of a waveguide? The cutback method measures transmission through several lengths of the same waveguide and takes the slope of loss against length, which separates propagation loss from coupling loss. Ring resonator Q gives the loss of short, low-loss waveguides more precisely. See cutback propagation loss and waveguide loss measurement methods compared.

Lab practice

How would you couple a free-space laser into a single-mode fiber? Match the focused spot to the fiber's mode field diameter by choosing the lens focal length, then align with two steering mirrors: peak the lateral position and focus, then walk the beam with both mirrors and iterate. A good answer also says what efficiency to expect: the product of the computed mode match, the lens transmission and the roughly 3.4% reflection at an uncoated fiber end, so that a shortfall can be recognized as misalignment or a poor beam. See free-space to single-mode fiber coupling.

How do you find the axes of a waveplate or set the polarization into a PM fiber? Between crossed polarizers, rotate the waveplate until the transmitted power is at a minimum; its axes are then aligned with the polarizers. For a polarization-maintaining fiber, align the input polarization to the slow axis and adjust it while the fiber is stressed or heated, minimizing the output's power variation, which maximizes the polarization extinction ratio. See measuring PER and PDL.

A measured fiber link shows 2 dB more loss than expected. What do you check? In order: connector end faces under a microscope (contamination is the most common cause; see fiber connector inspection and cleaning), the power meter's wavelength setting and reference, tight bends, and then an OTDR trace to locate the loss along the link. An interviewer is looking for a cheap-first, systematic order.

Preparing

Most of the calculations above use a handful of constants and formulas: $hc = 1239.8$ eV·nm, $P_\text{dBm} = 10\log_{10}(P/1\,\text{mW})$, the Fresnel reflection, the V number, the Gaussian beam's $z_R$ and divergence, the shot and thermal noise formulas and the FSR. Being able to apply them quickly with sensible typical values, and to say which assumption each answer depends on, covers much of what a technical screen asks. For the relationships between devices, types of lasers, types of optical fiber and bits to light are the site's overview pages.

References: B. E. A. Saleh and M. C. Teich, Fundamentals of Photonics, 3rd ed. (Wiley, 2019); G. P. Agrawal, Fiber-Optic Communication Systems, 5th ed. (Wiley, 2021).