Among IR detection noise, thermal noise holds a unique place by its ubiquity and fundamentality. For the small, low-cost, fast-start uncooled thermal camera, thermal noise is not only the main noise source but the core physical signature that distinguishes it from cooled detectors.
01 Physical origin: the never-resting dance of carriers
Thermal noise, also Johnson–Nyquist noise, originates in any conductor or semiconductor above absolute zero: carriers (electrons or holes) undergo random thermal motion, producing random voltage fluctuations across the material even with no applied voltage. Nyquist's theorem describes it precisely: V_n2 = 4k_B T R Δf (V_n RMS, k_B Boltzmann constant, T absolute temperature, R resistance, Δf electronic bandwidth).
- Universal: any resistive element above zero temperature has it; it cannot be eradicated;
- "White" noise: flat power spectral density over a very wide frequency range;
- Resistance-dependent: noise voltage grows with √R, but noise power is independent of R;
- Temperature-dependent: noise voltage is proportional to √T — the origin of the name and the only physical handle for "managing" it.
02 Its central role in the uncooled microbolometer
Mainstream uncooled detectors are microbolometers: a tiny pixel thermally isolated from the substrate by very long legs, with a high temperature-coefficient-of-resistance material whose resistance is temperature-sensitive. Thermal noise appears doubly: ① pixel-resistance thermal noise — the microbridge resistance R at temperature T produces an inherent thermal-noise voltage added directly to the signal; ② thermal-conductance fluctuation noise — unique to uncooled detectors, the microbridge exchanges heat with the substrate through thermal conductance G, a quantum-fluctuating process causing the bridge temperature to randomly wander, with equivalent temperature-noise power ΔT_n2 = (4k_B T2 Δf) / G.
G is decisive: small enough to give a clear temperature rise (high responsivity) when absorbing IR, but not so small that thermal-conductance fluctuation noise (∝ 1/√G) and the thermal time constant explode, hurting frame rate and stability. Uncooled design is therefore a high-wire act balancing responsivity, thermal noise, time constant and manufacturability — with thermal noise as the main background hum.
03 Counter-strategies: the ultimate parry with "body temperature"
- Lower operating temperature (limited): uncooled detectors sit near ambient (~300 K). A thermoelectric cooler holding or slightly lowering temperature (e.g. to 280 K) cuts thermal noise by √T, but TEC adds power, volume and cost with limited range.
- Optimize microbridge structure and material (main battlefield): thermal isolation via longer legs, smaller cross-section and low-conductivity materials (e.g. silicon nitride) raises responsivity but often also raises conductance-fluctuation noise — a trade-off; material engineering uses high-TCR materials (VOx, a-Si) and tunes pixel resistance R (tens of kΩ) to match the ROIC's optimal noise.
- ROIC and signal processing: low-noise pre-amplifiers matched to pixel resistance; correlated double sampling (CDS) to cancel pixel bias and some low-frequency thermal fixed-pattern noise; bandwidth limiting via low-pass filters to directly cut thermal-noise power (bounded by frame rate and response speed).
04 The technology roadmap through the thermal-noise lens
- Why NETD struggles below 50 mK? Today's best uncooled VOx sits at 30–50 mK, set by intrinsic material noise, conductance-fluctuation limits and silicon ROIC noise. Breaking through needs new materials (higher-TCR superlattice films) or new structures (extreme nanowire isolation).
- The fundamental gap vs cooled: cooled detectors (HgCdTe) chilled to 77 K or below suppress thermal noise exponentially (∝√T); limiting noise becomes photon or readout noise, easily reaching <10 mK or even 1 mK NETD. Cooling is the ultimate weapon but costs money, power and volume.
- Are high-operating-temperature (HOT) detectors the future? To cut cooler cost, the industry develops "hot" cooled detectors working at higher temperatures (e.g. 150–200 K). Their thermal noise is higher than at 77 K but still far below 300 K uncooled — a frontier balancing performance, cost and volume, whose core challenge is suppressing dark current and thermal noise at higher temperature.
Thermal noise, like the detector's own body-temperature murmur, draws a clear ceiling over uncooled IR technology. The fight against it drives the extreme of micro-nano processing, the search for new thermoelectric materials and relentless ROIC innovation.