Electronics Guide

Articles in This Category

Fundamentals of Photodetection

Detectors fall into two broad families. Photon detectors respond to individual quanta of light and dominate the ultraviolet through the short-wave infrared. Thermal detectors respond to absorbed radiant energy as heat, which makes their response nearly independent of wavelength but comparatively slow. Almost every detector described in this category belongs to the first family; microbolometers and pyroelectric sensors are the common exceptions.

Photon Absorption and Carrier Collection

Most photon detectors rely on the interaction between light and a semiconductor. When a photon with energy greater than the material's bandgap is absorbed, it promotes an electron from the valence band to the conduction band, creating an electron-hole pair. An electric field, usually supplied by a reverse-biased junction, sweeps these carriers apart so they can be collected as current, converting the optical signal into an electrical one. Photons whose energy falls below the bandgap pass through without generating carriers, which sets each material's long-wavelength cutoff. At the short-wavelength end, absorption becomes so strong that carriers are generated within the first few nanometers of the surface, where recombination can destroy them before collection; this is why ultraviolet response depends heavily on window material, passivation, and junction depth.

Figures of Merit

A consistent set of parameters describes detector performance and allows fair comparison across technologies.

• Quantum efficiency is the fraction of incident photons that produce collected carriers. Good silicon devices exceed 80 percent across much of the visible range, and back-illuminated scientific sensors approach 95 percent at their peak.
• Responsivity, expressed in amperes per watt, is the electrical output per unit optical input. For a fixed quantum efficiency it increases in proportion to wavelength, because each photon of longer wavelength carries less energy, so a given optical power delivers more photons. Silicon photodiodes peak near 0.5 to 0.6 amperes per watt around 900 nanometers, and InGaAs devices approach 1 ampere per watt at 1550 nanometers. Responsivity then collapses past the cutoff, where quantum efficiency falls to zero.
• Spectral response describes sensitivity versus wavelength and is set by the absorbing material together with any window, filter, or antireflection coating.
• Dark current is the output present with no illumination. It grows with temperature, junction area, and reverse bias; for silicon it roughly doubles for every 8 to 10 degrees Celsius of temperature rise, which is the physical basis for cooling sensitive detectors.
• Noise-equivalent power (NEP) is the optical power that yields a signal equal to the noise in a one-hertz bandwidth, stated in watts per root hertz.
• Specific detectivity (D*) normalizes NEP by the square root of active area and bandwidth, so detectors of different sizes can be compared directly. Its units, centimeters times root hertz per watt, are known as Jones.
• Response speed is governed by carrier transit time across the depletion region, by the RC time constant formed by junction capacitance and load resistance, and, in devices with an undepleted region, by slow carrier diffusion that produces a long tail on the impulse response.
• Linearity and dynamic range matter in radiometry and machine vision. A photodiode operated near zero bias into a virtual ground is linear over many decades, while gain-based detectors saturate as the multiplication region is depleted of carriers.

Photodiode Technologies

Photodiodes are the most widely used photodetectors, with several configurations tuned to different needs. PN-junction photodiodes offer simple, reliable construction for general light sensing. PIN photodiodes insert a lightly doped intrinsic layer between the p and n regions, widening the depletion zone to lower capacitance and improve both speed and quantum efficiency; they dominate fiber-optic receivers, where the fastest InGaAs devices reach bandwidths of several tens of gigahertz.

Avalanche photodiodes operate at high reverse bias, where impact ionization multiplies the photocurrent internally. That internal gain lifts the signal above the noise floor of the following amplifier, which is exactly what a weak optical signal needs. The gain comes at a price: multiplication is a statistical process, so it adds excess noise described by the excess noise factor, and the gain itself depends steeply on both bias and temperature. Practical linear-mode gains run from tens to a few hundred in silicon and roughly ten to twenty in InGaAs devices, and stable operation demands a regulated high-voltage supply with temperature compensation.

Biased above the breakdown voltage in Geiger mode, the same avalanche structure becomes a single-photon avalanche diode (SPAD). A single absorbed photon triggers a self-sustaining avalanche that a quenching resistor or active circuit must halt before the device can detect again, imposing a dead time of tens of nanoseconds. Because the output is a fixed-amplitude pulse, a SPAD reports arrival times rather than intensity. A silicon photomultiplier (SiPM) parallels hundreds or thousands of quenched SPAD microcells on one die, so the summed output becomes proportional to photon count over a limited range while retaining single-photon sensitivity.

Photodiodes also run in two distinct electrical modes. In photovoltaic mode the device sits at zero bias, typically across the virtual ground of an amplifier, which eliminates dark current from reverse leakage and gives the best linearity and lowest noise, though the large junction capacitance limits speed. In photoconductive mode a reverse bias widens the depletion region, cutting capacitance and transit time for high-speed work at the cost of added dark current and its associated shot noise.

Phototransistors combine detection with on-chip current gain, giving higher sensitivity than a bare photodiode but slower response and a less linear, temperature-dependent output. They are common in optocouplers and simple presence or threshold sensing rather than precision measurement. Photoconductors, including cadmium sulfide cells and infrared lead salt detectors, change resistance under illumination; they are inexpensive and simple to bias, but they respond slowly and drift with temperature and illumination history.

Spectral Range and Detector Materials

The choice of semiconductor determines the usable wavelength range, because the bandgap fixes the long-wavelength cutoff. Silicon responds from roughly 200 nm to 1100 nm, spanning the ultraviolet, visible, and near-infrared, and is the default for cameras, light meters, and short-reach optical links. Its combination of low cost, low dark current at room temperature, and compatibility with integrated readout electronics is difficult for any other material to match. Wide-bandgap materials such as silicon carbide and aluminum gallium nitride serve the opposite need: they are blind to visible light, which makes them well suited to flame detection and solar-blind ultraviolet sensing without heavy filtering.

Germanium and indium gallium arsenide (InGaAs) extend coverage into the short-wave infrared. Standard lattice-matched InGaAs detectors cut off near 1.7 µm and are the default receiver material for the 1310 nm and 1550 nm telecommunications bands. Extended-range variants push the cutoff to between roughly 1.9 µm and 2.6 µm for gas sensing, moisture measurement, and near-infrared spectroscopy, but the growing lattice mismatch with the indium phosphide substrate introduces defects and raises dark current, so these devices usually require cooling. Germanium covers a similar band at lower cost and carries substantially higher dark current, which has pushed it toward integrated silicon-photonic receivers rather than discrete instrumentation.

Beyond about 3 µm, mercury cadmium telluride (HgCdTe), quantum-well infrared photodetectors, and type-II InAs/GaSb superlattices serve the mid-wave and long-wave infrared bands used in thermal imaging and spectroscopy. HgCdTe is particularly valuable because adjusting the mercury-to-cadmium ratio tunes the cutoff across a wide range. All of these narrow-gap photon detectors generate carriers thermally as readily as optically, so they operate cooled, commonly near 77 K with a liquid-nitrogen dewar or a closed-cycle cryocooler.

Uncooled thermal imaging takes a different route. A microbolometer array absorbs long-wave infrared radiation on a thermally isolated membrane and measures the resulting temperature rise as a resistance change in a vanadium oxide or amorphous silicon thermistor. Because the mechanism is heating rather than band-to-band carrier generation, the response is broad and nearly flat with wavelength, and no cryogenic cooling is required. The trade-off is sensitivity and speed: commercial arrays with 12 to 17 µm pixel pitch reach noise-equivalent temperature differences on the order of 50 mK with thermal time constants near 10 ms, far slower than a cooled photon detector but adequate for handheld thermal cameras, building inspection, and automotive night vision.

From Single Elements to Arrays

Detectors are built in three formats, and the choice shapes the entire signal chain. A single-element detector measures total light on one active area and suits power meters, optical receivers, and photometry, where speed and linearity matter more than spatial detail. A linear array places hundreds or thousands of elements in a row and captures a one-dimensional profile in a single exposure; spectrometers use one to record an entire spectrum at once, and document scanners and web-inspection systems build a two-dimensional image by moving the object past the array. A two-dimensional array captures a full scene in one exposure and requires an addressing and readout architecture of its own.

Image Sensor Architectures

Image sensors place millions of photodetector elements, or pixels, in a two-dimensional grid. Charge-coupled devices (CCDs) accumulate charge in each pixel and shift it across the array to a single shared output amplifier, which historically yielded very uniform, low-noise images because every pixel passes through the same electronics. CMOS image sensors instead place amplification and addressing at each pixel. The now-standard four-transistor pixel pairs a pinned photodiode with a separate floating diffusion node, which permits correlated double sampling and pushes read noise to roughly one electron in scientific-grade devices. Placing the amplifier in the pixel also allows random access, higher frame rates, far lower power, and integration of timing, conversion, and processing on the same chip, so active-pixel CMOS sensors now dominate consumer imaging and have displaced CCDs in most scientific work.

Structural refinements have driven much of the recent progress. Back-illuminated sensors are thinned and lit from the rear so that metal wiring no longer blocks the light path, raising quantum efficiency at short wavelengths. Stacked designs bond the pixel array to a separate logic wafer, freeing area for per-column converters, memory, and on-chip processing. Production smartphone sensors now reach 200 megapixels at pixel pitches near 0.56 µm, which is comparable to the wavelength of green light and close to the limit set by diffraction and by the number of electrons such a small well can hold; makers recover usable low-light performance by binning neighboring pixels under a shared color filter. Specialized variants add high dynamic range through multiple exposures or dual conversion gain, global shutters that expose every row simultaneously to avoid the skew of rolling-shutter readout, and on-chip time-of-flight or event-based architectures that report per-pixel brightness changes instead of full frames.

Photomultipliers and Single-Photon Detection

When the signal amounts to only a handful of photons, gain must come before the first noisy amplifier, because no amount of later amplification can recover a signal already buried in the input stage's noise. Photomultiplier tubes solve this with vacuum electronics: a photocathode converts photons to electrons, and a chain of dynodes held at successively higher potentials multiplies each photoelectron by a factor of a million or more. The result is fast response, very large sensitive area, and negligible readout noise. The costs are bulk, fragility, sensitivity to magnetic fields, a supply of one to two kilovolts, and a photocathode whose quantum efficiency peaks at roughly 25 to 30 percent in the blue for common bialkali types.

Solid-state alternatives now cover much of the same ground. SiPMs and SPADs reach single-photon sensitivity from a bias of a few tens of volts in a rugged, compact package that tolerates magnetic fields, which is why they have taken over positron emission tomography detectors designed to operate inside magnetic resonance scanners. Their weakness is thermally generated dark counts, which are orders of magnitude more frequent than a photomultiplier's and are usually managed by cooling, by timing coincidence requirements, or by discarding counts below a multi-photon threshold.

Superconducting nanowire single-photon detectors occupy the extreme end of performance. A current-biased nanowire cooled to a few kelvin switches briefly out of the superconducting state when a photon breaks Cooper pairs, producing a voltage pulse. These devices deliver system detection efficiencies above 90 percent at 1550 nm, timing jitter measured in picoseconds, and dark count rates well below one per second, which makes them the reference detector for quantum key distribution, deep-space optical links, and photonic quantum computing. The cryogenic requirement keeps them out of routine instrumentation.

Microchannel plates and image intensifiers extend electron multiplication to two dimensions, amplifying faint scenes for night vision and for time-resolved scientific imaging. Electron-multiplying CCDs apply gain within the charge-shift register itself, and intensified CCD and CMOS cameras couple an intensifier to a conventional sensor, adding both gain and nanosecond gating for fluorescence lifetime and combustion studies.

Noise and Signal Conditioning

The smallest detectable signal is set by noise, not by responsivity alone. Several mechanisms contribute. Shot noise arises from the discrete arrival of photons and from the statistics of dark current, and it grows as the square root of the current, so a bright signal is easier to measure in relative terms than a faint one. Thermal, or Johnson, noise comes from resistive elements, notably the load or feedback resistor, and it falls as that resistance rises. Flicker noise dominates at low frequencies and pushes precision instruments toward modulated measurement. Excess noise accompanies avalanche gain, and read noise sets the floor for array sensors. In the best case a measurement becomes shot-noise limited, meaning the photon statistics themselves set the limit and no further electronic improvement helps.

Cooling attacks the dark-current term directly. A thermoelectric cooler holding a detector 30 to 40 degrees Celsius below ambient can cut dark current by roughly an order of magnitude, and cryogenic cooling is mandatory for narrow-gap infrared photon detectors.

Because most photodiodes behave as high-impedance current sources, the readout almost always begins with a transimpedance amplifier. Its feedback resistor sets the conversion gain in volts per ampere, and larger resistors give more signal for less amplifier noise, which is why high-sensitivity front ends favor gain resistors in the megohm to gigohm range. Bandwidth follows from the interaction of that resistor with the detector capacitance and the amplifier's own gain-bandwidth product, and a small feedback capacitor is normally required to keep the stage stable. This tension between sensitivity and speed is the central compromise in optical receiver design: reducing detector area cuts capacitance and buys bandwidth, but it also demands tighter optical alignment.

Practical performance depends as much on layout as on device selection. Photocurrents in the picoampere range demand guarded traces, clean grounding, electrostatic shielding, and heavily filtered bias supplies. Ambient light and 100 or 120 hertz ripple from artificial lighting are common interferers; modulating the source and recovering the signal with a lock-in amplifier or synchronous detector rejects both, along with much of the flicker noise. Optical filters, baffles, and careful mechanical design keep stray light from ever reaching the detector, which is usually cheaper than removing it electronically.

Selecting a Detector

Detector choice follows from a short sequence of questions, and answering them in order usually narrows the field to one or two candidates.

• What wavelength? The required band eliminates most materials immediately. Silicon serves the ultraviolet through the near-infrared, InGaAs the short-wave infrared, and cooled narrow-gap materials or microbolometers the thermal bands.
• How much light? Milliwatts call for a simple photodiode operated near zero bias. Nanowatts justify a cooled detector and a careful transimpedance front end. Individual photons demand a photomultiplier, SPAD, SiPM, or superconducting detector.
• How fast? Steady illumination allows a large-area, high-impedance design; gigahertz data rates force a small, reverse-biased PIN diode and an integrated receiver.
• Point, line, or scene? The measurement geometry decides between a single element, a linear array, and an image sensor, and that choice largely determines the cost and complexity of the readout.
• What accuracy? Radiometric work needs a calibrated, linear, temperature-stable detector; threshold detection or counting tolerates far less.
• What environment? Temperature range, magnetic fields, vibration, radiation, and available supply voltages routinely rule out the technically ideal device in favor of a robust one.

Applications

Light detection and sensing reaches nearly every industry. Consumer electronics depend on image sensors for cameras and on ambient-light and proximity sensors that dim displays and disable touchscreens during calls. Optical communication systems rely on PIN and avalanche photodiodes to recover data at rates now reaching hundreds of gigabits per second per wavelength. Scientific instruments use specialized detectors for spectroscopy, microscopy, and astronomy, where cooled, back-illuminated arrays integrate for hours on faint sources. Biomedical photonics applies optical sensing to pulse oximetry, optical coherence tomography, flow cytometry, and laboratory diagnostics. Industrial systems use photodetectors for machine vision and inspection, optical encoders that report shaft position, safety light curtains that guard machinery, and smoke detectors that watch for scattered or obstructed light.

Emerging needs continue to push detector design. Autonomous vehicles pair cameras with LIDAR and active optical sensing, which has driven rapid commercial development of SPAD and SiPM arrays with integrated timing circuitry. High-speed and event-based imaging feeds real-time computer-vision and machine-learning pipelines with far less data than conventional video. Quantum photonics depends on detectors that resolve individual photons with precise timing, and the reach of quantum key distribution and deep-space optical links is set largely by detector efficiency and dark counts.

Conclusion

Every light detector performs the same conversion, yet the engineering diverges sharply once wavelength, signal level, speed, and format are fixed. A photodiode and a transimpedance amplifier answer most measurement problems; an image sensor answers most imaging problems; and gain-based detectors answer the problems where photons are scarce. Understanding quantum efficiency, dark current, and the noise mechanisms that limit each device is what makes the choice between them systematic rather than arbitrary. The four subcategories above develop these device families in detail, and the companion category on light generation and sources covers the emitters that most of these detectors are paired with.