Package & Thermal Path in Reference Design

In practice, a reference’s accuracy and stability are set as much by its package and thermal path as by its bandgap topology. Once the reference leaves the datasheet test board and sits on a cramped, warm PCB, junction temperature and gradients start to rewrite the effective tempco and drift.

When you read tempco and drift numbers in a reference datasheet, you are looking at results measured on carefully controlled boards: modest power, well-defined copper areas and a friendly ambient. On a real product, especially with a tiny DFN sitting near hot power devices, the junction temperature can end up tens of degrees above those assumptions. The electrical specs have not changed, but the thermal context has, and that context decides how your reference behaves over time.

The same reference die can be sold in SOT-23, DFN or a small ceramic package. Electrically they look similar, but the thermal path is not: heat flows from die to package, into the leadframe or slug, through the solder pads and copper pours, across planes and finally into the enclosure and ambient. A small DFN on minimal pads, a SOT-23 on a generous copper island and a ceramic can on a thick board all produce very different junction temperatures and drift behaviour for the exact same die.

Designing the package and thermal path for a reference comes down to three decisions. First, choosing a package family sets the baseline θJA and θJC you can realistically expect on your board. Second, the power you burn in the device turns voltage headroom and load current into self-heating, which multiplies straight into tempco and drift. Third, pads, copper pours, vias and nearby hotspots shape where that heat actually flows, creating local thermal islands and gradients that can make two “identical” reference channels behave very differently.

Package Menu for Voltage/Current References

Package choice for a reference is rarely just about footprint and pinout. Each family comes with a different balance of thermal resistance, copper dependence and long-term stability, which in turn shapes how much of the datasheet accuracy you keep on the finished board. This section acts as a package menu, tying common families to their typical thermal behaviour and reference use cases.

SOT-23 / SC70 / SOT-223 Families

Small SOT and SC70 packages are easy to hand-solder and friendly for prototypes. With a modest copper island, their effective θJA can be quite reasonable, making them solid workhorses for moderate power and moderate accuracy references. Larger SOT-223 packages push more heat into the board and are often used when the reference or its surrounding circuitry must dissipate more power without resorting to DFN or QFN.

In industrial controllers or inverter gate-drive boards, a SOT-23 reference on a sensibly sized copper island is often the sweet spot between cost, layout flexibility and thermal control. It keeps routing straightforward while still allowing you to stretch θJA by growing copper where the layout permits.

DFN / MSOP / QFN Compact Packages

DFN, MSOP and QFN packages shrink the footprint and parasitics, and often include an exposed pad that can deliver very low thermal resistance when used correctly. The catch is that their θJA depends heavily on the pad pattern, copper area and via strategy. A tiny exposed pad that never connects to meaningful copper can trap the reference in a local hot spot even at modest power levels.

In PMICs, sensor modules and dense mixed-signal boards, a DFN reference keeps loops short and noise low, as long as you treat the exposed pad and vias as part of the thermal design. With a well-designed pad and copper pour it can outperform larger plastic packages; with a minimal pad it can become your worst thermal enemy.

SOIC / TSSOP Packages

SOIC and TSSOP packages occupy more board area but bring larger thermal mass and a more forgiving thermal path. Their θJA is typically well-behaved over a range of copper areas, and temperature changes on the board tend to translate into slower, smoother changes at the die. That makes them attractive for high-stability references on lab boards and industrial control backplanes.

Many classic buried-zener precision references still ship in SOIC-8 or MSOP-8 because these packages make it easier to achieve low noise, good long-term stability and predictable thermal behaviour over long mission times. They are also excellent choices for characterization boards and golden reference modules.

Ceramic and Metal Can Packages

Ceramic and metal packages sit at the premium end of the reference world. They cost more and consume more volume, but their materials and construction can withstand higher temperatures, aggressive environments and long mission durations with better control of drift. They are common in aerospace, military and very high temperature modules where plastic packages would age or shift too quickly.

For applications that push well beyond 125 °C or demand exceptionally low long-term drift, ceramic packages help the reference survive where standard plastic parts struggle. They are rarely needed in mainstream industrial or consumer designs but become the default in extreme environments.

θJA, θJC, θJB and Junction Estimation

Thermal metrics such as θ_JA, θ_JC and θ_JB look abstract in a datasheet, but each one is just degrees Celsius per watt between two points in the heat path. Once you know which two points they connect, you can turn them into quick estimates of junction temperature and decide whether a given package and layout are safe for your reference.

θ_JA is the junction-to-ambient thermal resistance, describing how many degrees the die warms up for each watt dissipated when heat flows into the surrounding environment. θ_JC is junction-to-case, usually to the top or bottom of the package as defined in the datasheet, and θ_JB is junction-to-board, measured at a point on the PCB near the device. Most datasheets quote these values on standard test boards; your own PCB, copper areas and enclosure will shift the effective numbers up or down.

A practical way to use these metrics is to work backwards from power. First, estimate the power dissipated in the reference from supply current, headroom voltage and output load. Second, multiply that power by a realistic θ_JA for your layout to get an approximate junction rise, ΔT ≈ P × θ_JA. Third, add that rise to the expected ambient or board temperature, plus any extra from nearby hot devices, to see where the die will really sit during operation. This does not need to be perfect; even a rough estimate will show whether you are a couple of degrees above ambient or ten degrees hotter than you thought.

For example, a 15 mW reference in a small package with θ_JA around 150 °C/W only gains about 2–3 °C of self-heating, which is easy to absorb in most error budgets. The same device on a cramped, poorly cooled board can run much warmer if the effective θ_JA is higher or if hotspot coupling raises the local temperature. What matters is less the exact θ_JA value and more how it changes with copper area, pad patterns and vias: parts that are very sensitive to board conditions demand more careful layout if you want predictable reference behaviour.

Self-Heating, ΔT and Vref Drift

Once you know the thermal resistance of your reference and its mounting, every milliwatt of power you burn becomes a predictable temperature rise. That rise multiplies through the tempco and long-term drift of the device, often consuming a visible part of the error budget in high-temperature or automotive environments.

Power in a reference comes from more than the supply current listing in the datasheet. The internal bandgap, amplifiers and trimming networks draw quiescent current from the supply, and any headroom between VIN and the reference output turns that current into heat. Output loads and external resistor networks add their own contribution. A simple first estimate of dissipation is P ≈ (VIN − VOUT) × I_total, plus a small term for any current that flows through the reference output into the rest of the circuit.

With an estimate of P and a realistic θ_JA, you can approximate self-heating as ΔT ≈ P × θ_JA. On a cool industrial board at 25 °C, a 20 mW reference with θ_JA around 120 °C/W will see only about 2–3 °C of temperature rise, so the junction might sit near 27–28 °C in steady state. In a compact automotive module at 85 °C, a 40 mW reference in a small package with θ_JA closer to 150 °C/W can see a 6 °C rise or more, and local hotspots may push the die towards 95–100 °C under worst-case conditions.

Translating ΔT into drift is straightforward. If a reference has a tempco of 10 ppm/°C at 2.5 V, a 6 °C rise adds about 10 × 10⁻⁶ × 6 × 2.5 V ≈ 150 µV, or 0.06 mV of shift. At 25 ppm/°C, the same 6 °C rise becomes roughly 0.15 mV. On a bench this may be negligible; in a high-accuracy rail or automotive safety function it can consume a noticeable slice of the allowed error. The crucial point is that self-heating is a deterministic offset, not random noise, so it must be accounted for explicitly in any serious error budget.

By combining a simple P × θ_JA estimate with your tempco numbers, you can quickly see whether self-heating is a minor correction or a dominant term that demands changes to package, layout or operating conditions.

PCB Thermal Path: Pads, Copper and Planes

For a precision reference, pad size, copper pours, vias and planes are part of the thermal design, not just routing details. Changing them alters how heat escapes from the die and how strongly the reference is coupled to the rest of the board, which can shift junction temperature by several degrees and quietly eat into drift and accuracy.

A small pad under the reference keeps the thermal contact area limited. The device is less tied to large planes, so it may be somewhat insulated from far-away hotspots, but its own heat has fewer paths to escape and the local temperature rise is higher. Enlarging the pad and adding a copper pour reduces local thermal resistance, lowering self-heating, yet it also ties the reference more tightly to whatever temperature that copper region runs at, especially when it belongs to a busy ground plane.

Connecting the pad directly into a broad ground plane helps spread heat and makes the board temperature more uniform, which can be very helpful in quiet areas. Near power devices, the same connection can back-feed heat from hot copper into the reference and push its junction temperature close to the worst-case power corner. Using a narrow neck or thermal relief pads reduces that coupling, at the cost of slower heat flow out of the device. A good layout balances these two extremes so that the reference shares enough copper to avoid a local hotspot but is not welded into the hottest region of the plane.

Vias are another steering handle for the thermal path. A handful of vias under or near the pad can pull heat into inner or backside planes and flatten gradients, while too many vias into a high-power copper area may drag the reference up to the same temperature as MOSFETs, regulators and shunt resistors. On a simple two-layer board most heat spreads laterally across the top copper and through the board material, whereas on a four-layer board inner planes can carry heat over a much larger area. For a reference, the goal is to give it a reasonably sized copper island and via path into a relatively quiet plane, instead of tying it directly into the hottest current returns.

Thinking of pads, pours, vias and planes as parts of a single thermal path makes it easier to choose geometries that support your accuracy targets: enough copper and connectivity to control self-heating, but not so much that the reference simply becomes a thermometer for the nearest power stage.

Local Hotspots, Gradients and Channel Mismatch

Board temperature is never a single number. Local hotspots from power devices and vertical gradients to enclosures create zones at very different temperatures. Multi-channel references and redundant rails that were tightly matched on the bench can show much larger differences once each channel sits in its own thermal micro-environment.

Horizontally across a board, power stages, inductors and motor drivers raise the temperature in their corner while connector, I/O and digital control areas remain cooler. It is common to see 15–20 °C spread between a hot power corner and a quiet region. A multi-channel reference that promises a few ppm per degree of internal matching can easily show tens of ppm of channel-to-channel drift if one channel consistently runs near 65 °C and another sits closer to 45 °C on the same PCB.

Vertical gradients add another layer of variation. Parts mounted closer to a metal enclosure or heatsink may run cooler or hotter, depending on how that structure is used, while components facing an air gap rely mostly on convection. If one reference or channel is on the side pressed against a warm housing and another is buried deeper in the assembly, their junction temperatures can differ by several degrees even when the measured ambient looks similar at a distance.

In practice, this means internal matching specifications only tell part of the story. Channel mismatch at board level is set by both the IC design and the thermal landscape of the PCB. Grouping critical channels in a region with similar thermal exposure, keeping redundant references in comparable locations and avoiding having the most important rail closest to the hottest device all help preserve matching. For the very tightest channels, a clean, shared thermal environment is often more important than shaving a few millimetres from the routing.

Treating the board as a temperature map rather than a single ambient point makes it easier to understand why some reference channels drift more than others and where layout changes can flatten gradients and improve long-term consistency.

Mission Thermal Profile: Ambient, Enclosure and Application

A reference IC never lives on the PCB alone. Ambient temperature, enclosure design, mounting position and airflow together create a mission thermal profile that can push the junction tens of degrees away from the datasheet conditions. Understanding this profile is just as important as picking the right package and PCB thermal path.

On an open bench the environment is gentle. Room temperature sits near 25 °C, the board is exposed to free convection and the only self-heating comes from the components themselves. In this case the effective junction-to- ambient thermal resistance for a precision reference, such as a 2.5 V ADR4525 in SOIC-8, is often close to the values quoted for standard test boards. Bench measurements are useful for characterising the intrinsic device, but they do not represent cabinets, modules or outdoor enclosures.

Inside industrial cabinets the whole baseline shifts. Control enclosures frequently run at 40–60 °C with poor airflow, multiple hot boards stacked and local hotspots around power converters. Even with a careful copper island and via pattern, a reference like ADR4525 on a mid-layer board may see a junction temperature 20–30 °C higher than in bench tests. Cabinet layout, board spacing and fan placement become part of the thermal design for the reference.

Automotive under-hood modules push things further. Ambient around the module can sit at 85 °C during normal hot operation and climb towards 125 °C in worst-case conditions. Compact AEC-Q qualified references such as 2.5 V SGM40242Q in SOT-23 or small DFN packages live inside tightly packed ECUs, surrounded by switches, drivers and DC/DC converters. The board couples thermally to an aluminium housing or bracket, and in many designs a thermal pad or gel connects the board to the metal shell. Depending on how the module is mounted, the enclosure may act as a cooler or as a large heat source that tracks engine or inverter temperature.

Outdoor and field equipment introduce solar loading and weather. A white cabinet in direct sun can reach 70–80 °C on its outer surface even when ambient air is only 35 °C. Dark plastic housings can run hotter still. If the enclosure is sealed for IP rating, internal air circulation is poor and the board largely follows the wall temperature. A simple NTC probe glued to the housing, combined with an on-board temperature sensor near the reference, is often the only realistic way to build an accurate picture of the long-term thermal environment.

Between the reference package and the outside world there is always a second thermal path. The package body exchanges heat with air, potting compounds, thermal pads and nearby metal, while the PCB side conducts heat into mounting screws, brackets and chassis. In a benchtop design free convection dominates; in sealed automotive and outdoor modules solid conduction through the housing can take over, sometimes pushing heat into the reference instead of removing it. No matter how well the PCB is optimised, an enclosure that clamps a hot metal wall against the reference area will quickly erase most of the benefits.

Thinking in terms of mission profiles – bench, cabinet, under-hood, outdoor – helps you decide whether the reference needs extra thermal isolation, closer monitoring or a different package altogether, long before field drift and unexpected error growth show up in production.

Validation: Measuring Temperature and Drift on the Bench

Once the package, PCB and enclosure are chosen, the only way to know whether the reference behaves as expected is to measure it on real hardware. A simple test setup with a few temperature sensors and controlled power profiles can reveal how junction temperature and drift behave across your mission thermal profile.

A practical validation setup starts with representative reference ICs and local temperature sensing. Precision devices such as ADR4525BRZ and automotive-grade references like SGM40242Q can be placed in the same footprint options to compare behaviour. A high-accuracy digital thermometer (for example a TMP116 in a small DFN package) or a TMP100-Q1 in SOT-23 allows you to monitor the PCB temperature within a few millimetres of the reference. One or two 10 kΩ NTCs such as NCP18XH103J03RB soldered next to the package give an independent analogue view, while a leaded NTC probe bonded to the housing can track enclosure temperature.

The first validation step is an ambient sweep. Place the assembled board in a small chamber or insulated box and step the environment across key points such as −20, 0, 25, 60, 85 and 105 °C. At each step, allow the board to settle and then log Vref, the nearby digital sensor reading and the NTC resistance. Plotting Vref against the measured board temperature reveals the effective board-level tempco, which can then be compared with the device datasheet and folded back into your error budget.

Next, characterise self-heating by sweeping power. Adjust supply voltage, load current and any resistor networks so that the reference dissipation moves from a few milliwatts up to the worst-case value in your design. For each operating point, record the change in local temperature reported by the sensors and the corresponding shift in Vref. This directly validates the P × θJA ≈ ΔT estimate and shows whether self-heating remains a small correction or becomes a major contributor to drift in your layout.

Finally, perform a hotspot coupling test. Place a controllable power stage, such as a MOSFET and shunt resistor, near the reference and toggle it between low and high dissipation while logging Vref and temperature over time. Watch for both fast steps and slow ramps in the reference output as copper and the board warm up. This reveals whether nearby power devices introduce unacceptable transient offsets or long-term drift that were not visible in static ambient sweeps.

The data from these tests can be summarised in a simple matrix comparing layouts, packages and enclosures. Folding the measured tempco, self-heating and hotspot response back into the error budget closes the loop between theory and hardware and provides concrete design rules for future boards.

BOM & Procurement Notes for Package and Thermal Path

In a precision reference design, the package and thermal path are part of the specification, not an afterthought. If the BOM only lists voltage, accuracy and temperature range, a second source can easily introduce a very different θJA or footprint and quietly break the drift and stability budget you designed for.

The same 2.5 V reference die in a roomy SOIC-8 with a solid copper island behaves very differently from a tiny DFN soldered into a cramped under-hood module. Junction temperature, self-heating and coupling to hot planes are all set by the package family, pad pattern and enclosure. A robust BOM makes these constraints explicit so they are not lost during sourcing.

Recommended thermal-related fields in the BOM and specification

• Package family and size: specify the package clearly, for example “SOIC-8 3.9 × 4.9 mm, no exposed pad” or “DFN-8 3 × 3 mm with exposed pad”. Avoid vague “8-pin” descriptions that allow radically different thermal behaviour.
• Board-level thermal condition or θJA class: if it matters, state a target such as “θ_JA ≤ 120 °C/W on 4-layer board with ground plane under device and copper island ≥ 50 mm²” or reference an internal thermal class that suppliers must meet.
• Maximum allowed dissipation and self-heating: describe limits in system terms, for example “Max dissipation 30 mW at 85 °C ambient; design target ΔT_junction ≤ 5 °C”. This tells layout and second sources how much thermal headroom is left in the error budget.
• Environment and case temperature range: specify realistic conditions such as “Module case up to 95 °C; local board near reference up to 110 °C” or “Outdoor cabinet wall up to 80 °C in sun, internal air 60 °C”. Suppliers then know this is a genuinely high-temperature application, not a room-temperature instrument.

Example part numbers and why their package and thermal behaviour matter

A precision 2.5 V reference such as ADR4525BRZ in SOIC-8 is easy to cool with a modest copper island and a few vias. The same family in a 3 × 3 mm DFN needs careful exposed-pad layout to reach comparable θJA. If the BOM only says “2.5 V, 0.02 % reference”, a DFN variant or another vendor’s DFN device may be dropped in without thermal re-qualification.

In automotive modules, a part like SGM40242Q in SOT-23 looks attractive for space, but its tiny footprint and proximity to hot drivers mean θJA is strongly layout-dependent. A second source with a different SOT-23 pad recommendation or a larger MSOP/DFN alternative will not behave the same thermally, even if voltage, accuracy and tempco match within the datasheet tables.

The thermal sensing chain also deserves explicit part numbers. A high-accuracy digital sensor such as TMP116, a car-grade TMP100-Q1 near the reference and 10 kΩ NTCs like NCP18XH103J03RB or leaded NXFT15 probes on the housing give you the data needed to validate θJA and self-heating. If these are loosely specified as “any 10 kΩ NTC”, calibration and drift compensation quickly lose meaning.

Second-source selection: beyond electrical compatibility

When qualifying a second source, require more than matching voltage and tolerance. Check that the package family, outline and recommended pad pattern are compatible with the existing PCB, and that the vendor’s θJA data on a comparable 4-layer board is within a defined margin. For AEC-Q applications, treat any change from SOT-23 to DFN or exposed-pad MSOP as a new thermal design that needs lab validation.

A simple rule helps procurement: if a candidate part changes the package family, pad style or thermal class, it is not a “drop-in replacement” and must go through the same self-heating and mission-profile tests as the original. Capture this rule in the BOM notes so that sourcing and layout teams understand where substitutions are safe and where they are not.

Package and Thermal Path FAQs for References

This FAQ answers practical questions about how package choice and thermal paths affect real reference accuracy. Use these short answers as a quick checklist when you size copper, place hot parts and define BOM rules, so your voltage and current references stay stable from prototype to volume production.