An Electron Multiplier (EM) is a high‑sensitivity vacuum photodetector based on the secondary electron emission effect. Featuring current gain above \(10^8\), nanosecond response speed, low noise and high counting rate, it is widely applied in online VOCs monitoring for environmental protection, mass spectrometry in food safety, single‑molecule fluorescence detection in life sciences, nuclear radiation detection, vacuum leak testing, and ultraviolet / extreme‑ultraviolet photon detection. Photon counting mode represents its core application for ultra‑weak signal detection, enabling single‑photon‑level measurement. The high‑voltage bias power supply serves as an essential supporting unit for EM photon counting systems, providing high‑precision and high‑stability bias voltage to the multiplier stages. Its long‑term stability, temperature drift, low‑noise performance and linearity directly determine the EM’s gain stability, counting accuracy, detection limit and dynamic range. In photon counting mode, the EM operates within a high gain range of \(10^5\)~\(10^7\), where gain changes exponentially with bias voltage: a mere 1% voltage fluctuation may cause over 10% gain variation. Therefore, extremely demanding criteria are imposed on the power supply: 1. Ultimate voltage stability: long‑term stability better than ±0.05%/8 h and short‑term stability better than ±0.02%/min; otherwise severe gain fluctuation leads to counting errors, missed or false photon events, and failure in single‑photon detection. 2. Ultra‑low temperature drift: the EM gain itself is temperature‑sensitive; the power supply’s temperature coefficient must be ≤ 5 ppm/℃ to prevent significant counting deviation during quantitative analysis under varying ambient temperatures. 3. Ultra‑low output noise and ripple: peak‑to‑peak ripple < 10 mV and noise density < 20 nV/√Hz@1 kHz; otherwise amplified noise drastically raises dark counts, submerges weak photon signals and reduces sensitivity. 4. Excellent linearity and load regulation: load regulation better than ±0.05% to avoid gain drop, counting saturation and nonlinearity at high counting rates when anode current fluctuates across multiple orders of magnitude. Traditional high‑voltage power supplies suffer from large temperature drift, poor long‑term stability, high load regulation and excessive noise, failing to satisfy photon counting requirements. Designs must comply with standards including GB/T 4793.1 *Safety requirements for electrical equipment for measurement, control and laboratory use — Part 1: General requirements* and JJF 1164‑2006 *Calibration specification for bench GC‑MS instruments*, while meeting high‑precision and high‑stability demands for ultra‑weak quantitative detection. Targeting the core requirements and technical challenges of EM high‑voltage bias power supplies for photon counting, this methodology establishes a full‑process general framework covering high‑stability topology, full temperature‑range compensation, gain closed‑loop control, low‑noise optimization and high‑linearity output. It adapts to Channel Electron Multipliers (CEMs) and continuous dynode electron multipliers, providing standardized design guidelines for domestic precision instruments in environmental monitoring, mass spectrometry and life science research. Adopting the main architecture of **low‑noise resonant boost + high‑voltage linear regulation + fully digital dual closed‑loop control**, combined with full temperature‑range compensation and online gain calibration, this solution eliminates traditional drawbacks, achieves ppm‑level temperature stability and outstanding long‑term gain consistency, and fully meets quantitative photon counting demands. The design follows five core principles: First, a two‑stage topology of **front‑end quasi‑resonant flyback boosting + rear‑end high‑voltage linear regulation** balances efficiency, ultra‑low noise and high stability. The front quasi‑resonant inverter realizes zero‑voltage switching to minimize switching noise and loss, achieves electrical isolation and high efficiency above 92%, and restricts output ripple below 20 mV to supply clean input for the rear stage. The rear low‑dropout high‑voltage linear regulator eliminates all switching ripple and residual noise with no switching action, delivering ultimate low noise and stability. Parallel power transistors reduce on‑resistance, enhance transient response under dynamic load current and maintain high linearity across wide operating conditions. Second, full temperature‑range drift suppression ensures stable EM gain via three layers: low‑drift component selection, hardware compensation and algorithmic full‑temperature modeling. Critical components employ buried Zener references with TC ≤ 2 ppm/℃, high‑precision metal‑foil sampling resistors ≤ 1 ppm/℃, low‑leakage SiC Schottky rectifiers and low‑TC polypropylene film capacitors to minimize inherent thermal drift. Passive temperature compensation networks counteract offset drift in references and amplifiers, limiting hardware TC ≤ 3 ppm/℃ within 0 ℃~50 ℃. During factory calibration, multi‑point sampling across −10 ℃~60 ℃ establishes polynomial drift models stored in memory. In real operation, high‑precision temperature sensors feed thermal data to dynamically adjust DAC references and PID parameters, ultimately reducing the overall system temperature coefficient to ≤ 2 ppm/℃ and eliminating temperature‑induced gain variation. Third, closed‑loop gain stabilization and online calibration address EM aging over lifetime. Anode current feedback compares measured anode current with theoretical target gain, dynamically adjusting bias voltage to compensate gain decay and maintain constant effective gain. Built‑in standard light sources enable automatic reference calibration using known photon flux to recalculate actual gain and correct high voltage without system shutdown or disassembly. Aging prediction models integrate cumulative operating time, counting statistics, high voltage and temperature data to pre‑compensate long‑term gain fading and guarantee counting accuracy throughout the device lifecycle. Fourth, full‑link low‑noise and high‑linearity optimization suppress noise at the source, enhance filtering and improve load transient performance. SiC MOSFETs and SiC Schottky diodes reduce switching noise; low‑noise high‑PSRR error amplifiers with input noise ≤ 5 nV/√Hz@1 kHz strengthen ripple rejection; three‑stage cascaded RC low‑pass filtering further suppress wideband interference, achieving final ripple < 5 mV and noise density < 15 nV/√Hz@1 kHz to control dark counts. Optimized loop bandwidth and low‑impedance output ensure load regulation ≤ ±0.03% and line regulation ≤ ±0.02%, with transient response < 10 μs during abrupt current changes, preventing gain drop and nonlinearity at high counting rates. Full‑range multi‑point voltage calibration corrects system nonlinearity to ensure output linearity better than ±0.01% for reliable quantitative analysis. Fifth, a fully digital high‑precision control platform uses FPGA+ARM dual core: ARM manages human‑machine interaction, temperature compensation and calibration logic; FPGA executes high‑speed closed‑loop regulation, data acquisition and protection. A 24‑bit high‑resolution ADC samples voltage and current at 100 kHz; optimized digital PID with loop bandwidth ≥ 10 kHz guarantees fast, stable and accurate regulation. Digital filtering stabilizes sampled data to avoid control jitter caused by noise. All calibration tables, thermal models and parameters are stored in non‑volatile memory, supporting flexible configuration for different EM models and gain characteristics. Reliability and safety protection serve as critical supporting measures. Multi‑level hardware‑software dual protection includes input under/overvoltage, output overvoltage, overcurrent/short‑circuit, overtemperature, high‑voltage spark discharge and vacuum interlock, with response < 1 μs for instant shutdown and fast capacitor discharge to protect expensive multipliers from surge damage. Current limiting prevents accelerated aging under excessive anode current; soft start avoids voltage inrush during power‑on. All key components adopt Class I derating with voltage ≤70%, current ≤60% and temperature ≤80% of rated values; long‑life non‑electrolytic high‑voltage capacitors and full conduction thermal design ensure uniform temperature distribution and extended service life, achieving an MTBF ≥ 50,000 hours for continuous operation in monitoring stations and laboratories. Strict insulation safety compliance provides reinforced isolation withstandoff exceeding twice the maximum output voltage; fully enclosed high‑voltage assemblies feature power interlock and discharge protection for operator safety, with real‑time monitoring, fault alarms and log recording for maintenance traceability. In summary, this methodology forms a complete technical framework covering high‑stability topology, full temperature compensation, gain closed‑loop control and low‑noise high‑linearity optimization, resolving traditional weaknesses such as large thermal drift, poor stability, high load regulation and excessive noise. It realizes temperature coefficient ≤ 2 ppm/℃ via combined hardware‑software compensation, lifelong constant gain via closed‑loop calibration, and ultra‑low ripple below 5 mV via two‑stage regulation and multi‑stage filtering. Fully compatible with EM photon counting requirements for ultra‑weak quantitative detection, it is widely applicable to VOCs monitoring, mass spectrometry, single‑molecule studies and nuclear detection, delivering core technical support for performance upgrading and domestic independent development of high‑end precision scientific instruments.