NETD, pixel pitch, and responsivity are the core parameters on an infrared detector datasheet. Starting from physical definitions and mathematical relations, this article explains what determines each — the thermosensitive material, microbridge thermal design, and MEMS processing — and how the three couple in a finished system.
1. Pixel structure
Each pixel of an uncooled detector is a suspended microbridge: an infrared absorber sits on two low-conductance silicon-nitride legs, separated from the readout integrated circuit (ROIC) below by a vacuum gap, all sealed inside a vacuum cavity. Incoming infrared radiation is converted into heat by the absorber; the bridge warms up; the resistance of the thermosensitive film changes accordingly; and the ROIC converts that change into a voltage signal. Each parameter traces back to one link: the temperature rise comes from thermal design, the precision of the resistance measurement from materials and electronics, and the achievable size from MEMS processing.
2. NETD
NETD (noise equivalent temperature difference) is the smallest temperature difference a detector can resolve, in mK — the smaller the value, the higher the thermal sensitivity. In weak-contrast scenes its engineering meaning is: SNR is roughly proportional to the temperature difference divided by NETD. For an early overheat defect on a transformer joint with a temperature rise of only 0.1 °C (100 mK), a detector with NETD = 80 mK yields SNR ≈ 1.25 and the hot spot is difficult to separate from noise; at NETD = 40 mK, SNR ≈ 2.5 and the hot-spot outline becomes discernible.
NETD is defined as the ratio of total output noise to responsivity:
NETD = Vn / R
R is the temperature responsivity (the derivative of output voltage with respect to scene temperature difference), and Vn is the total output noise, summed in power from mutually uncorrelated stages:
Vn = √( σ²tf + σ²roic + σ²adc + … )
σtf is the microbridge temperature-fluctuation noise, σroic the ROIC noise, and σadc the ADC quantization noise. NETD is not an intrinsic device constant; its reported value is bound to test conditions: radiation reaching the pixel scales inversely with the square of the lens F-number — the same device measures nearly twice the NETD under an f/1.4 lens compared with f/1.0 — so datasheets state “@25 °C, f/1.0”.
In the denominator, R originates from the temperature coefficient of resistance (TCR) of the thermosensitive film — the relative rate of resistance change with temperature. Mainstream vanadium oxide (VOx) films have a TCR of about −2% to −3%/K at room temperature; amorphous silicon (α-Si) is around −2%/K. TCR is controlled by film composition and phase: VOx mixed-valence films are deposited by reactive sputtering, and composition drift changes TCR and noise together, making material consistency a key process control point in detector manufacturing.
The noise terms in the numerator have distinct origins: temperature-fluctuation noise is governed by thermal conductance and heat capacity, and the ROIC contributes Johnson and 1/f noise. Thermal conductance G is the key variable of microbridge thermal design: thinner, longer legs and lower material conductance give a larger temperature rise for the same absorbed power; vacuum packaging removes the gas-conduction term, and any loss of vacuum degrades NETD.
Device parameters separate into two layers: process and materials fix the device physically — once TCR, thermal conductance, and readout noise are set, they are intrinsic to that device; NETD and responsivity are ultimately obtained by measurement. The common engineering reference is with lens attached at F/1.0: radiation collected through an f/1.0 optic, measured against a blackbody target at the stated temperature, defines the product specification. The frame rate is part of the measurement conditions as well: it sets the integration time, and the same device measures differently at 25 Hz versus 50 Hz, so datasheets state the frame rate alongside temperature and F-number (e.g. @f/1.0, 300 K, 30 Hz). Values measured on this reference can differ from the detector’s factory report (bare die-level test) — lens transmission and alignment enter the measurement chain, so confirm a common measurement basis when comparing specifications.
Typical NETD requirements by application: security surveillance ≤50 mK, industrial temperature measurement ≤60 mK, outdoor observation and power-line inspection ≤40 mK, scientific and medical ≤30 mK.
3. Pixel pitch
Pixel pitch is the center-to-center distance between adjacent pixels, in μm. The mainstream today is 17μm and 12μm: at the same 640×512 resolution, a 12μm chip has nearly half the area of a 17μm one.
The direct gains from a smaller pixel are area and optics, not heat capacity. At constant resolution the die area scales with the square of pitch: 12 μm is about (12/17)² ≈ 0.50 of 17 μm. At fixed F-number and field of view, focal length and aperture scale by 12/17 ≈ 0.71. Shutter area follows the sensor. Steady-state bias power is set by bias voltage, pixel resistance, and pulse duty cycle; heat capacity changes τth, not that power. Lower dissipation has to be designed into the bias timing and the readout.
The difficulties concentrate in process, materials, and noise. Pixel area scales with the square of pitch: shrinking from 17μm to 12μm halves the radiation each pixel receives — converting the weaker signal into a valid digital output while suppressing noise is the central problem of small pixels. Photolithography line width and overlay define the smallest microbridge; sacrificial-layer release forms the suspended structure for thermal isolation; bump and bond interconnect pitch sets a floor under the pixel pitch; and TCR and noise characteristics must stay consistent on ever-smaller films. Raytron announced an 8 μm 1920×1080 array in 2021 and a 6 μm 640×512 array in 2024. Its 2025 statement was that the 8 μm series had entered volume introduction, while 6 μm volume production still depends on process maturity and demand. An announcement is not a shipping status.
Rated parameters of several representative uncooled detectors:
Table 1. Rated NETD of representative uncooled detectors
| Detector | Pitch | Array | Rated NETD | Notes |
|---|---|---|---|---|
| Lynred PICO640 Gen2 | 17 μm | 640×480 | <50 mK | f/1.0, 300 K, 30 Hz |
| Lynred PICO640S | 17 μm | 640×480 | <30 mK | same conditions; thermal time constant 12 ms |
| Global Sensor Technology GST612C | 12 μm | 640×512 | typ. <35 mK | ceramic packaging |
| Global Sensor Technology GST612W | 12 μm | 640×512 | typ. <40 mK | wafer-level packaging |
| Teledyne FLIR Boson+ core | 12 μm | 640×512 | ≤20 mK | camera-core NETD, including lens and processing; not comparable with a bare detector |
The scaling route has a physical boundary. In the LWIR band (8–14μm), the diffraction blur spot at f/1.0 is about 25μm across — the same scale as a 12μm pixel; as pitch shrinks further, the spot grows relative to the pixel and the resolution gains from optics taper off. Past 12μm, the limiting factors extend beyond line width to how closely packaging, readout, and algorithms can approach the diffraction limit.
4. Responsivity
Responsivity measures how efficiently the detector converts infrared radiation into an electrical signal, commonly in mV/K. VOx microbolometers typically deliver about 10–30 mV/K under typical bias, f/1.0 optics, and integration time. The voltage change is approximated by:
ΔV ≈ Ib · R0 · TCR · ΔT
The bias current Ib flows through the pixel resistance R0; a scene temperature difference raises the bridge temperature by ΔT (proportional to absorbed power over thermal conductance); TCR converts that rise into a resistance change, and the output voltage follows. Responsivity is not an independently adjustable parameter but the combined output of material TCR, bias conditions, and microbridge thermal design, which is why its rated value is stated together with bias voltage, F-number, and integration time.
Thermal design introduces the thermal time constant:
τth = C / G
The ratio of heat capacity C to thermal conductance G sets the speed of the bridge’s temperature response: smaller G gives higher sensitivity but slower response; τth must be well below the frame period (milliseconds to tens of milliseconds, matching 30–60 Hz frame rates), otherwise moving targets smear. Sensitivity and response speed trade off on the same thermal structure — the fundamental trade-off of microbridge design.
The cost of excessive responsivity is dynamic range: responsivity acts as signal-chain gain, and viewing hot targets (sunlit metal or sky, for example) easily saturates the chain and loses detail in bright regions. What matters in evaluating responsivity is not the number itself but its match with dynamic range and the backend ADC range.
NETD and responsivity have to be read together, and the symbol must not change meaning mid-page. The R used above is scene-temperature responsivity in mV/K. F-number, transmission, and pixel area are already inside it:
NETD = Vn / R_T
Doubling R_T and halving Vn move NETD by the same factor. They are not the same operation. The first spends bias and integration time and changes self-heating and dynamic range; the second is a readout and process problem. Integration time touches both: a longer integration narrows the noise bandwidth and drops the frame rate. Once a UAV or vehicle fixes the frame rate, only bias and the circuit remain.
To take the lens back out of the responsivity, use power responsivity ℜ in V/W. The change in power on the pixel scales as τ·A·ΔT / (4·F#²), so
R_T ∝ ℜ · τ · A / (4·F#²), NETD ∝ 4·F#² / (τ · A · ℜ)
ℜ is not the mV/K quantity. Using one symbol for both divides F-number and pixel area twice. Before comparing two NETD numbers, match F-number, frame rate, and blackbody temperature.
Conclusion
NETD maps to material TCR and noise control, pixel pitch to MEMS processing and diffraction physics, and responsivity to thermal design and electronics. The three parameters constrain one another and together define the design space of the final system — principles determine the parameters; the parameters define the boundaries.
