Thermal & Fan Control for Power Supplies and Adapters
← Back to: Power Supplies & Adapters
This page explains how to design multipoint temperature sensing, fan and pump control, derating curves and telemetry so that PSU and adapter thermal behaviour stays predictable, safe and quiet over the full operating range.
It focuses on practical placement, control policies and IC building blocks that turn cooling from a single NTC and fan into a controlled, diagnosable thermal management system.
Thermal and fan control in power supplies and adapters is not just an “OTP or no OTP” question. Real systems must balance hotspot reliability, audible noise, fan and pump lifetime, user experience and regulatory temperature limits across very different form factors.
In server and CRPS or ATX supplies, multiple fans and long airflow paths create hidden hotspots around SR MOSFETs or magnetics that a single ambient NTC cannot see. In compact adapters, fast chargers and LED drivers, enclosure temperature and touch feel matter as much as internal silicon limits. Industrial 24 V front-ends and medical PSUs must remain predictable under dusty enclosures, elevated cabinet temperatures and strict acoustic and safety constraints.
Relying on a single temperature sensor feeding a hard over-temperature shutdown leads to abrupt power loss, noisy fans running harder than necessary and components quietly operating above their intended junction limits. Without explicit derating curves, the PSU behaviour at high ambient temperature becomes unpredictable for system designers and operators.
This page focuses on multipoint temperature sensing, fan and pump actuation and thermal derating policies in power supplies and adapters. Primary over-voltage, over-current and short-circuit protection and hot-swap SOA protection are handled in the dedicated protection and eFuse/hot-swap pages.
A power supply’s thermal map is defined by where losses are generated, how heat flows through copper, heatsinks and the enclosure and how air moves from inlet to outlet. Hotspots are typically found around the PFC bridge and choke, primary switches, LLC or flyback transformers, SR MOSFETs, output inductors and bulk capacitors along the airflow path.
In a 1U or CRPS form factor, long, narrow PCBs and line-shaped airflow mean that components near the outlet often run much warmer than those near the inlet. In compact adapters and LED drivers, dense placement and limited convection create strong local gradients between internal hotspot silicon and the plastic enclosure surface. In in-car accessory PSUs, cold-crank start-up and elevated cabin temperatures stretch the allowable operating range while airflow is often constrained by trim parts and harnesses.
A useful thermal map separates three layers: internal component junction and case temperatures that determine reliability; enclosure or chassis temperatures that determine user touch limits and safety; and ambient air temperature, which defines the boundary condition for derating. Multipoint temperature sensing, fan placement and control and power derating curves all need to be referenced back to this map.
This section focuses on where to think about temperature in the system and how it interacts with the digital controller and PMBus telemetry. Detailed over-voltage, over-current and short-circuit protection policies are covered on the dedicated protection page and are not repeated here.
A modern power supply rarely relies on a single temperature sensor. Multipoint sensing combines low-cost NTCs, higher-accuracy RTDs, junction-based sensing and integrated or digital temperature sensors so that thermal control decisions are based on both hotspot and system-level information instead of a single ambient point.
NTC thermistors attached to heatsinks, transformers or copper pours are typically used as protection and control points because of their cost, size and good thermal coupling. RTDs are reserved for reference locations that require better linearity and long-term stability. Junction-based sensors inside controllers or power ICs give direct visibility of silicon temperatures, while digital I²C or SMBus sensors provide convenient ambient, outlet or enclosure readings without additional analog front-ends.
Placement strategy starts by classifying each location as a protection point, a control point or a logging point. Protection points sit close to the most thermally stressed components and feed over-temperature shutdown or hard derating decisions. Control points represent the overall thermal state and drive fan curves or gradual power derating. Logging points are used for monitoring and predictive maintenance, tracking temperatures around components such as electrolytic capacitors, connectors or specific cabinet zones without directly influencing control loops.
Single-fan supplies can often be covered by a small number of sensors: a protection point near the worst-case hotspot, a control point that represents average thermal stress and possibly one extra logging point. Multi-fan or multi-module systems tend to require sensors per airflow region or per module, with control logic based on the maximum or a weighted combination of these readings to avoid blind spots along the airflow path or between parallel power modules.
Each additional sensor improves visibility but increases cost, routing complexity and susceptibility to noise. Thermal designers therefore prioritise the few locations that most strongly influence lifetime and user experience, ensure robust thermal coupling and avoid routing sensor traces close to high dv/dt nodes or magnetics. This section focuses on which temperature points to measure and how to classify their roles; ADC selection and detailed error budgeting are handled in separate sensing-focused content.
Once the sensing points are defined, the next step is to connect each temperature element to an ADC or digital interface without sacrificing accuracy or robustness. NTC thermistors typically form a resistive divider, RTDs are driven by a current source or bridge circuit, junction-based sensors feed matched front-ends and digital temperature sensors communicate via I²C or SMBus to the thermal controller or system host.
For NTC dividers, the series resistor is selected so that the most relevant temperature range occupies a useful span of the ADC input, and the divider is referenced to a stable supply or reference voltage. Simple RC filtering at the ADC input helps to reject switching noise from PWM stages, while the firmware compensates the non-linear NTC curve via look-up tables or polynomial approximations. RTD front-ends rely on precise excitation currents and often differential measurement paths to cancel lead resistance for long runs between sensor and controller.
Long sensor wiring, such as enclosure or cabinet probes, introduces cable resistance and additional EMI pickup. Shielded or twisted-pair cabling, careful routing away from high dv/dt nodes and suitable filtering are required to keep readings stable. When digital temperature sensors are used, they move the sensitive analog conversion close to the sensing point and send digital data back over I²C or SMBus, reducing analog error but imposing constraints on bus routing, pull-up values and signal integrity.
Multiple channels can be acquired using dedicated multi-channel sigma-delta converters, a multiplexer feeding a SAR ADC or integrated ADCs inside a digital PSU controller. Sigma-delta converters favour higher resolution and noise rejection at moderate sampling rates, which suits temperature and slow telemetry. Multiplexer plus SAR ADC schemes trade some complexity in channel sequencing and settling time for lower cost. Integrated controller ADCs reduce component count but require careful planning of channel allocation across voltage, current and temperature inputs.
Thermal control loops do not require microsecond response: fan speed and derating typically update on time scales from tens of milliseconds to seconds, whereas temperature protection logic must still respond quickly enough to prevent runaway under cooling failures. Absolute accuracy, repeatability and worst-case margins matter more than raw resolution. Detailed current and voltage sensing architectures share similar ADC resources and are covered in the dedicated Current/Voltage Sensing page; this section concentrates on the temperature side of the interface.
Cooling hardware in power supplies ranges from simple 2-wire DC fans to 4-wire PWM fans with tachometer feedback and liquid-cooling pumps. A suitable driver architecture must match each load type while protecting the 12 V or 24 V auxiliary rail from inrush, managing acoustic noise and enabling reliable fault detection for stalled or disconnected devices.
Two-wire fans rely on voltage or low-side or high-side PWM modulation and provide no direct speed feedback. Three-wire fans add a tachometer output so that RPM can be monitored and compared against the requested speed. Four-wire fans separate the DC supply from a dedicated logic-level PWM input and usually operate at high PWM frequencies to reduce audible artefacts, while still exporting tach pulses for diagnostics. Blowers and axial fans differ in pressure and flow characteristics, but are driven using the same basic electrical interfaces.
Pumps and liquid-cooling loops bring similar requirements with higher inertia and often higher continuous current. Depending on system architecture, pumps may be treated as high-current fan loads driven by robust low-side or high-side switches, or as separate BLDC motors controlled by specialised pump driver ICs with integrated commutation, soft-start and locked-rotor protection. In all cases, the driver must coexist with the rest of the auxiliary rails without injecting excessive ripple or interference.
Fan and pump driver topologies include low-side MOSFET switches, high-side switches, linear control and PWM-based modulation. Single-channel drivers suit compact adapters and small industrial PSUs, while multi-channel fan driver ICs consolidate multiple outputs, tach inputs and fault reporting for server and telecom supplies. Tachometer measurements and current-sense information enable detection of stalled rotors, missing loads and wiring faults, feeding fault flags into the thermal controller and, where applicable, PMBus status registers.
Soft-start and current limiting on fan and pump rails prevent start-up surge from collapsing 12 V or 24 V auxiliary supplies when several loads start simultaneously. Driver ICs therefore incorporate programmable ramp profiles, sequencing and per-channel fault timers so that a stuck or shorted fan is isolated without bringing down the entire PSU. This section focuses on low-power cooling drivers; high-voltage bus hot-swap, inrush limiting and SOA enforcement for 400 V or main DC rails are covered in the dedicated eFuse & Hot-Swap content.
Thermal derating policies translate measured temperatures into fan or pump speed and ultimately into allowed output power and current. A clear derating strategy avoids abrupt shutdowns, exposes predictable behaviour to system designers and aligns PSU operation with datasheet ratings such as “100 % load at 40 °C, linear derating above 50 °C”.
Linear derating policies map temperature above a threshold into a proportional reduction in capability, such as reducing available power by a fixed percentage per degree above a reference point. Segment-based schemes define separate regions for normal operation, warning, derating and over-temperature protection. Each region can include different fan-speed targets, power limits and reporting behaviour so that the system is warned before hard protection thresholds are reached.
Derating curves also depend on environmental and operating conditions. Higher altitude reduces air density and effective cooling, so a design may rely on more conservative power limits or earlier derating thresholds above a defined altitude. Low input voltage, such as 90 Vac operation for universal AC front-ends, can increase RMS currents and loss, requiring different derating behaviour compared with nominal mains. Fan or pump faults reduce cooling capacity and should trigger immediate transition into a stricter derating mode rather than waiting for temperature to reach normal OTP levels.
Implementation can be fully analog, using NTC-based networks and comparators to realise thresholds and piece-wise slopes, but digital control using MCU or digital PSU controllers enables more flexible lookup tables and piece-wise linear functions. Firmware can blend multiple variables—temperature, altitude setting, input voltage range and fan health status—into a single power limit and fan-speed command. PMBus or similar interfaces then expose the current derating state, available power and thermal warnings to the system host for logging and power budgeting.
Different applications expect different derating behaviour. Server and telecom PSUs prioritise continuous operation with gradual power reduction and rich telemetry instead of sudden shutdown. Adapters and fast chargers need smooth transitions between fast-charge and reduced power modes without oscillation, while LED drivers favour soft brightness reduction over flicker or blackouts. Datasheet derating graphs summarise these policies in terms of ambient temperature and sometimes altitude; inside the PSU, those curves are enforced by the combination of thermal sensing, fan and pump drivers and digital or analog control logic.
Thermal safety in power supplies focuses on how reduced cooling capacity and rising temperatures are handled before hardware reaches absolute limits. Fan or pump failures, blocked airflow and harsh ambient conditions all need a structured failsafe path that reduces output capability, raises clear alarms and only resorts to shutdown when controlled derating can no longer keep hotspot temperatures within safe margins.
When a fan or pump fault is detected through tachometer loss, abnormal current or health flags, the thermal controller should first switch into a conservative derating mode. Maximum output power or current is reduced and remaining cooling devices are driven to higher speed or flow. If temperatures continue to climb in spite of derating, a timed window of limited operation provides a controlled transition to shutdown instead of an abrupt OTP event. This staged response gives connected systems time to migrate workloads or secure processes.
Redundant cooling, such as N+1 fan arrays or dual pumps, allows operation to continue through single device failures but does not remove the need for limits. Policies define how much power and for how long a PSU may run with one or more failed cooling channels, depending on ambient temperature and load class. In mild conditions, loss of a single fan may be acceptable with little or no derating; in hot racks or sealed cabinets the same failure should trigger immediate power reduction and a clear maintenance request to avoid cumulative damage and unplanned trips.
Thermal safety also interacts with other protection functions. As hotspot temperatures approach predefined limits, over-current and peak current thresholds can be tightened so that short overload events do not produce additional heating on already stressed components. Over-temperature protection thresholds must be set with margin between measured sensor points and actual junction temperatures, accounting for thermal resistance, transient gradients, sensor accuracy and ageing. This ensures device junction limits are never crossed even in worst-case combinations of ambient and load.
Each thermal safety action should be logged so that throttling episodes, OTP occurrences and fan or pump failures are visible during field diagnostics and return analysis. The detailed configuration of over-voltage, over-current and short-circuit comparators remains the responsibility of the dedicated OV/OC/SCP Protection stage; this section focuses on how thermal control cooperates with those blocks to maintain safe, predictable behaviour under cooling stress.
Thermal telemetry turns internal temperatures, cooling states and derating decisions into data that can be monitored, logged and compared across power supplies. Real-time visibility of hotspot temperatures, inlet and outlet readings, fan and pump speeds, command duty cycles and derating levels enables system designers and operators to understand how a PSU behaves in the field instead of treating thermal behaviour as a black box.
Useful telemetry extends beyond instantaneous values. Logs for thermal throttling episodes, over-temperature events and fan or pump failures provide context for preventive maintenance and return analysis. Counters for the number of times derating has been engaged, total time spent in each thermal region and the history of OTP triggers help identify installations that operate close to limits, suffer from blocked airflow or have cooling hardware that is degrading faster than expected.
These signals can be exported through several interfaces. PMBus or SMBus offers a rich digital channel with standard commands for temperatures, fan speeds and status bits, along with manufacturer-specific registers for derating level, fan command duty and event counters. Simpler systems can use GPIO pins to expose thermal warnings, derating states and fan-fail flags, while legacy or analogue environments can receive a voltage output representing a key temperature or derating index for connection to PLC inputs or simple monitoring circuits.
In server and data-center power systems, thermal telemetry is consumed by DCIM and orchestration tools that compare PSU behaviour across racks and positions. Supplies that frequently reach high fan speeds, derating states or near-OTP temperatures can be flagged for cleaning, load redistribution or replacement long before failures occur. In industrial and infrastructure applications, telemetry passing through gateways to SCADA or cloud platforms supports similar analysis, highlighting cabinets with inadequate ventilation or unusual ambient conditions and guiding planned maintenance windows instead of reactive shutdowns.
Designing these capabilities in from the start requires reserving ADC channels for key temperature and current measurements, allocating register space for real-time values and counters, and using consistent naming aligned with PMBus or other industry conventions. Standardised thermal telemetry fields allow different PSU designs to be compared and integrated into common dashboards, making thermal performance, derating behaviour and cooling reliability visible across fleets instead of only during bench testing.
System-level questions
Thermal sensing, cooling control and safety checklist
Telemetry, logging and integration checklist
IC role mapping (example device classes)
The following device classes illustrate how thermal sensing, cooling control and telemetry functions can be partitioned into IC roles. Device names are examples for reference and do not restrict supplier choice.
What this page solves
Thermal map and system context in PSUs
Multipoint temperature sensing & placement strategies
Temperature sensing front-ends & ADC integration
Fan and pump driver architectures for PSUs
Derating curves and control policies
Safety, failsafe and interaction with protection functions
Telemetry, logging & field diagnostics
Design checklist & IC role mapping
Digital temperature monitor with multiple remote diodes or NTC inputs and SMBus or I²C interface, often with programmable limits and alarm outputs.
Example device classes: TMP451-class, TMP461-class, ADT7481-class, EMC1403-class.
ADC or monitoring frontend for NTC dividers, RTD bridges and auxiliary voltages, providing 12–16 bit resolution and digital readout.
Example device classes: LTC2997-class, INA219/INA226-class current and voltage monitors, similar multi-input monitor ICs.
2/3/4-wire fan controller with PWM outputs, tach inputs, soft-start and configurable fan curves for one or more channels.
Example device classes: EMC2302/EMC2305-class, MAX31790-class, NCT3941-class fan drivers.
Intelligent power switch for 12 V or 24 V rails with over-current protection, diagnostics and optional current sensing.
Example device classes: TPS1Hxxx-class automotive high-side switches, AOZ1xxx-class protected switches.
Integrated driver for small BLDC pumps or blowers with commutation, soft-start and locked-rotor protection.
Example device classes: DRV10866-class, DRV10xx-class, MP6530/MP6540-class BLDC drivers.
Controller that combines ADC channels, PWM or timer outputs and PMBus or I²C, implementing fan curves, derating tables, protection coordination and logging.
Example device classes: MSP430F5xx-class or MSPM0-class MCUs, STM32G4-class or STM32F3-class mixed-signal MCUs, dsPIC33-class digital power controllers, UCD3138-class digital PSU controllers.
Digital isolators and interface bridges that allow thermal telemetry and PMBus traffic to cross isolation boundaries while maintaining safety ratings.
Example device classes: ISO1640-class or ISO1540-class I²C isolators, ADuM1250-class digital isolators, standard USB–PMBus lab adapters for development.
Request a Quote
Thermal & Fan Control – FAQs
1. How many temperature sensors does a PSU really need and where should they be placed?
A practical design places at least one sensor on the main inlet, one near the hottest power devices and one on the outlet or case. Larger PSUs benefit from extra sensors on PFC, LLC or SR stages and magnetics. Some points drive protection, some drive fan control and others only serve logging and diagnostics.
2. What is the difference between control temperature points and protection temperature points in a PSU?
Control temperature points feed fan curves and derating algorithms and are allowed to move within a wide band during normal operation. Protection temperature points are reserved for absolute limits and OTP trips, with extra margin to guarantee device junction limits are never exceeded. Mixing both roles on one sensor often compromises safety or performance.
3. How accurate do PSU temperature measurements need to be for reliable derating curves?
Derating does not require laboratory-grade accuracy, but total error should be smaller than the desired step size and safety margin. For example, when using 5 °C derating steps, combined sensor, ADC and layout error ideally stays within about ±2 °C. Repeatability and noise filtering often matter more than absolute accuracy for stable behaviour.
4. When should remote digital temperature sensors be used instead of simple NTC dividers?
Remote digital sensors are attractive when long traces, strong switching noise or tight accuracy requirements make simple NTC dividers difficult to stabilise. Server and telecom PSUs with many hotspots, high airflow and strict derating guarantees often use digital monitors, while small adapters and LED drivers usually tolerate lower cost NTC and ADC solutions.
5. How should multi-fan and pump architectures be planned for 1U servers versus compact adapters?
Slim 1U or CRPS supplies benefit from multiple smaller fans arranged along the airflow path, often with N+1 redundancy and independent monitoring. Compact adapters usually rely on conduction and natural convection, with at most one small fan or pump. Architecture is shaped by enclosure size, airflow path, noise limits and required availability under fault conditions.
6. How can fan speed control be tuned to balance acoustic noise, lifetime and thermal margin?
A robust curve keeps fans slow and quiet at low temperature but ramps decisively when hotspots approach predefined thresholds. Hysteresis and time filtering avoid constant small adjustments. Lifetime benefits from avoiding unnecessary high speed, while thermal margin demands fast response near limits. Curves should be validated in real enclosures rather than only on open benches.
7. How should thermal derating curves be defined and documented so that users know what to expect?
Derating curves should clearly show output capability versus ambient or inlet temperature, highlighting regions for full power, partial derating and shutdown. Separate curves for different input voltages, airflow conditions or fan-fail modes avoid surprises. Curves need to be based on tested hardware, not optimistic calculations, and should be consistent with certification statements and marketing claims.
8. What is a practical strategy when a cooling device fails but the system still needs to keep running for a while?
A practical strategy immediately reduces power limits and drives remaining fans or pumps harder, then monitors hotspot rise with a defined timer. If temperatures stabilise below safety limits, the PSU can run in a degraded state. If not, a controlled shutdown after the timer expires protects hardware while giving the system a window to react gracefully.
9. How should fan-fail and NTC open or short faults be handled and verified end-to-end?
Faults such as fan stall, missing tach pulses or NTC open and short should trigger reliable detection, entry into a conservative derating mode and, if temperatures still rise, a timed shutdown. Designs benefit from explicit test cases that inject each fault, record thermal behaviour and confirm that alarms, power reduction and shutdown follow the documented policy.
10. Which thermal telemetry signals give the most value to DCIM or SCADA systems?
The most useful signals are inlet temperature, one or more hotspot temperatures, per-fan RPM and command duty, current derating level and thermal alarm or OTP flags. Counters for time spent in derating and the number of fan-fail and OTP events help DCIM or SCADA software identify stressed supplies and schedule preventive maintenance before outages occur.
11. Which IC building blocks are typically used to implement multipoint sensing and fan control in PSUs?
Multipoint thermal control usually combines multi-channel temperature sensors or monitor AFEs, dedicated fan controller ICs or protected high-side switches, and a microcontroller or digital PSU controller with ADC, PWM and PMBus or I²C. Digital isolators may separate telemetry from primary-side domains. Exact partitioning depends on channel count, safety requirements, firmware complexity and cost targets.
12. What are the most common thermal and fan-control mistakes that cause PSU field issues?
Common mistakes include relying on a single case NTC to represent all hotspots, omitting derating curves and only using hard OTP, ignoring fan-fail or NTC fault handling, and underestimating airflow blockage in real enclosures. Missing telemetry and logging also make root-cause analysis difficult, turning intermittent thermal stress into premature failures and unexplained field returns.
