Electrical Drive Considerations
Constant Current vs. Constant Voltage Drive
The characterization of thermoelectric modules is commonly done on a constant current basis; some specified amount of current is passed through a device under test, and the resulting voltage appearing across the device and thermal performance curves are reported for some specified hot-side temperature. Citing the wider availability and/or perceived ease of use of power supplies with fixed-voltage outputs, many may ask whether it is possible to drive thermoelectric devices from a constant-voltage source.
The simple answer to that question is yes; electrically speaking, TEMs behave like a sloppy resistor, with a more or less linear relationship between applied voltage and current flow. The relationship between the two varies with temperature difference across the device, but does not exhibit diode-like nonlinearities that compel use of constant-current supplies or other ballasting techniques.
Looking more closely however, shows that the two approaches do yield somewhat different behaviors. Looking to the voltage vs. temperature plots for the CP85438, it can be seen that for a constant current input, the voltage across the device increases with increasing ΔT. Consequently, as ΔT increases, so does the electrical power applied to the device. For a 5.1A input for example at a ΔT of zero, the electrical input power is about 48W, rising to about 60W at a ΔT of 55°C. By comparison, applying a fixed 12V to the device would result in about 6.2A of current flow at zero ΔT for 74W of electrical power input, decreasing to the same 60W, 12V operating point at ΔT=55°C.
If the system balances out thermally at a 55°C ΔT, there would be essentially zero difference at steady-state between using a 12V constant-voltage source and a 5.1A constant-current supply; the operating point works out the same either way. Because the fixed-voltage case results in application of greater power to the TEM when ΔT is near-zero however (as it would be when the system is initially started for example) it is likely that the fixed-voltage case would reach the steady-state condition more quickly than a constant-current source chosen to give the same steady-state operating point. Similarly, a system powered in constant-voltage fashion will tend to see a reduction in power applied to the TEM in response to system changes that push things toward a higher ΔT, such as dust accumulation on a heat sink. This results in a more self-limiting behavior less inclined to overheat than one driven in constant-current fashion.
In a nutshell, driving a TEM using a constant voltage versus a constant current can both work, but the resulting system behavior is slightly different. The distinction is most relevant in cases where one selects a fixed-value supply, connects it, and accepts whatever performance results; one relies on the intersection of the supply and TEM operating curves to determine an operating point in that case. If one instead uses some sort of control system to adjust the electrical input to the TEM, the control system takes over as the dominant factor in determining an operating point, with the TEM and supply characteristics simply determining the envelope within which the control system can function.
Figure 1: CP85438 operating curves. At a ΔT of approximately 55°C, a 12V constant-voltage input and a 5.1A constant-current drive are essentially equivalent.
Applications requiring maintenance of a constant temperature generally use either a thermostatic or a linear control scheme. In the first case, a fixed-value electrical input to the device is applied in “on/off” fashion in order to maintain temperature within some limited range of the desired setpoint. In the second, the electrical input is adjusted over a continuous, variable range between zero and some maximum value, in order to match the cooling effect produced to the thermal load on the system.
Thermostatic control has the advantages of being relatively simple and inexpensive to implement, but for reasons of reliability it is not preferred. The on/off cycling characteristic of this control scheme causes the thermoelectric device to experience repeated temperature swings of significant magnitude, and the resulting thermal expansion and contraction in the assembly causes mechanical stress and fatigue in the thermoelectric device which can lead to premature failures.
Linear control techniques tend to be more costly to implement due to the need for a variable power source, but are preferred for reliability reasons due to the lesser degree of thermo-mechanical stress they place on the TEM. Their implementation is also more complex; careful design and modelling of a thermal system is needed in order to design a feedback & control mechanism that allows maintenance of a target temperature without oscillations or instability. In particular, the potential for increased electrical input to a TEM to result in a reduced net cooling effect can complicate efforts at control system design.
Regardless of the control technique used, there is a potential for things such as fan failures, airflow obstructions, or other unexpected influences to cause a TEM used for cooling to overheat and thereby fail and/or cause damage to other equipment. They’re commonly assembled using solder-based techniques, so if one (or both) sides of a device reach the point where the solder used to assemble the device melts, it’s possible for a device to fail in an open- or short-circuit manner or simply fall apart. For this reason, provisions to monitor the hot-side temperature of a TEM and disconnect power if it exceeds a safe limit are a necessary component of a safe and reliable thermoelectric system.
Figure 2: An illustration of linear versus thermostatic control techniques
Electrical Supply Waveforms
Many approaches exist for obtaining a DC power source of varying effective amplitude, many of which might be better described as “unipolar” rather than “DC” insofar as the output produced may be highly time-variant, despite not changing polarity. Viewed over a long time span, a thermostatic control scheme like that described above would be an example. On shorter time scales, various schemes may involve waveforms with significant AC ripple content at frequencies anywhere from a few to a few million cycles per second.
The thermal mass of a TEM structure and attached equipment is typically large enough that significant temperature changes cannot occur when ripple in the drive waveform has a frequency in cycle-per-second territory or greater. As a result, high-frequency ripple does not cause the same sort of thermo-mechanical stress problems associated with thermostatic (on/off) control, which typically operates on time scales of tens of seconds or longer and depends on detection of substantial temperature changes in the system for its operation.
Such ripple content does however have an adverse effect on system efficiency. The heat pumping effect provided by a TEM is generally proportional to the current flow through the device, while the electrical input power applied to the device is proportional to the square of current flow. To a loose first approximation, doubling the current flow and operating for half the time results in the same amount of thermal transport through the cold side of the device, but consumes twice the amount of electrical energy per unit of thermal energy pumped in order to achieve that similar end result. In practice this approximation doesn’t hold exactly as there are some factors it doesn’t take into account, but it does offer insight as to why use of drive waveforms with high levels of ripple is not recommended.
Figure 3: Several drive voltage waveforms providing comparable time-averaged power input to a TEM, and their relative favorability.
Using the CP85438 and assuming a hot side temperature of 50°C and ΔT of 40°C for example, the difference between operating at a constant 3.4A or 6.8A with a 50% duty cycle (yielding the same average drive current) can be seen to be quite significant. Though a slightly higher average Q is predicted for the latter case, the coefficient of performance (average input power per watt of cold-side heat transport) falls from about 0.51 to 0.36 if a direct PWM control was used. The use of a simple electrical filter to convert a PWM waveform to its DC equivalent in this case could be expected to increase electrical efficiency by roughly 40%.
The significance of this effect will change with operating conditions, actual supply waveforms, and other factors. Generally speaking, maintaining ripple levels of 10% or less in the drive waveforms applied to a TEM is suggested for best results. Less is better, and direct full-scale (unfiltered) PWM control is not recommended. Similarly, rectified AC waveforms without filtering are not preferred, though they might be suggested or encountered due to the convenience of simply attaching a bridge rectifier to the output of a mains-fed transformer.
It should be noted that the performance of series-configured modules is more sensitive to ripple in the applied drive waveforms, due to the compounding nature of inefficiencies at each stage. In such cases, limiting ripple to 1% or less may be desirable.
Figure 4: Operating points for CP85438 at 40°CΔT, Th=50°C, at drive currents of 3.4A and 6.8A.
Mechanical Application Considerations
Mechanical properties of TEMs
Typical thermoelectric modules used for cooling applications are constructed of bismuth telluride-based semiconductor pellets sandwiched between thin sheets of an aluminum oxide-based ceramic. The pellets are electrically interconnected by solder joints to metallic bridge strips that are bonded to the ceramic plates. Since both the semiconductor and ceramic materials are rather brittle, a TEM is quite strong in compression across its thickness, but easily damaged by shear or bending forces. Tensile loading, while tolerated to a degree, acts as if to dismantle a TEM and separate it from its external heat transfer surfaces, rather then holding the lot together. Since any yielding to such tensile forces implies damage or degradation of the system, it’s not preferred.
Figure 5: Maintaining a compressive mechanical force (left) on a TEM is preferred. Bending (center) or shear (right) loading is more likely to result in damage or failure.
The means by which a thermoelectric device is physically interfaced with its surrounding environment has significant influence on thermal performance and the mechanical durability of the assembly. Adhesive bonding, mechanical clamping, and soldering are the three commonly used methods, which offer various advantages and disadvantages.
An adhesively bonded thermoelectric device is essentially glued in place using a thin layer of a bonding agent, usually an epoxy modified for improved thermal conductivity. This fixation method is permanent, and does not allow disassembly for maintenance, repair, or rework. It is also rigid, not freely allowing for relative motion between the thermoelectric device and the heat sinks/spreaders/etc. to which they are attached. This results in transfer of mechanical stresses between them, caused by differences in material expansion rates and temperature differentials as a system warms and cools. The magnitude of such stresses increases with device size, leaving adhesive bonding better suited for smaller thermoelectric devices; use of devices no larger than about 20mm (0.8in) on a side has been suggested both for adhesive- and solder-bonded applications. Relative to other methods, the thermal conductivity of this mounting method is modest and may vary over time to different extents depending on the character of the bonding agent used; products specifically characterized for their tolerance to the thermal and mechanical stresses involved are recommended over general-purpose adhesives for this reason. Most adhesives will also release some amount of vapor during the curing process, and afterward also if subject to a strong vacuum. This has potential to weaken the adhesive bond and/or introduce unwanted substances into an evacuated region. Finally, adhesive bonding by itself relies on the thermoelectric device as a structural member of the overall assembly, requiring that it support mechanical loads resulting from shock or vibration. Since most thermoelectric devices are relatively brittle, the potential for damage to result from such physical insults is significant, to a growing degree as the mass of system components supported by the TEM is increased.
Figure 6: Aluminum, the typical material from which heatsinks are made, expands at roughly 3 times the rate of the aluminum oxide from which the heat transfer surfaces of TEMs are commonly made. A rigid bond between the two such as is formed by an adhesive or soldered bond thus results in a set of opposing mechanical forces as an assembly of the two increases in temperature (right). A reciprocal set of forces is created on the cold side of the assembly, if a rigid bond is present there also.
Thermoelectric devices manufactured with a metallized outer surface can be fastened using a soldered connection. This assembly method is similar to adhesive bonding in its attributes, creating a rigid bond best suited for small-format devices. The bond produced tends to have better thermal conductivity however and is better suited to use in vacuum environments due to a lack of outgassing, while also allowing limited rework/repair potential due to the solder’s ability to be re-melted. Careful temperature control and solder material selection is needed when using this assembly method however, as the internal elements of the device are typically also assembled using a soldering process and can be damaged by excessive temperatures during assembly.
Figure 7: A TEM with a metallized surface suitable for soldering (right) versus one without (left)
Use of external fasteners or spring tension to compress a thermoelectric device between its hot- and cold-side thermal reservoirs is a common and recommended mounting method for larger-format devices. Typically used in conjunction with a thermal grease or similar thermal transfer aid, the lack of a rigid bond between system elements relieves mechanical stresses caused by thermal expansion. The external pressure provided by the fasteners helps to keep the mechanical forces applied to a TEM within an assembly compressive in nature, making feasible the use of the relatively massive heat sinks/spreaders necessary for dealing with the thermal loads generated by larger thermoelectric devices. As with adhesive bonding, the thermal interface materials used in clamped applications have a tendency to outgas under vacuum, which is problematic for some applications. Since no permanent mechanical bond is created, mechanically fastened assemblies can be disassembled for repair, maintenance, or rework with relative ease.
Because the mechanical fasteners used in a TEM assembly represent a thermal leakage path, apply the clamping force needed to maintain a low interface thermal resistance, and are subject to wide variations in temperature, their selection and application is a matter of some importance. Suggested practices include:
- Using stainless steel fasteners
- Using the minimum feasible fastener diameter
- Using thermally-insulating shoulder washers to create a thermal break between fastener and fastened item
- Using Belleville-type spring washers to maintain clamping force as thermal expansion & contraction cause dimensional changes in the assembly
Figure 8: A cross-section view of a suggested fastener system for mechanically-fastened thermoelectric assemblies. Use of a fiber shoulder washer to create a thermal break between fastener and cold/hot plate reduces thermal leakage, while use of a spring-type washer helps maintain clamping force despite dimensional changes caused by thermal expansion effects.
Fastener placement is also a consideration. Though rigid to the touch, heat sinks are flexible and can bend under clamping pressure, often to a surprising degree. Positioning fasteners along an axis that is parallel with any longitudinal fins on the heatsink used allows the fin structure to contribute to the structure’s stiffness, whereas positioning fasteners along a line perpendicular to the fins tends to allow a much higher degree of flexure.
Suggested clamping pressures for TEM assemblies are in the range of 150-300PSI (1-2N/mm2). For a square TEM 40mm on a side, the upper limit of this range results in a force comparable to the tensile strength of (2) typical #4-40 machine screws made of 18-8 stainless steel. Because torsional forces are also applied during assembly and the fastened elements will typically be made of comparatively-soft aluminum, a pair of #6 or M4 fasteners is suggested as a reasonable starting point for fastener selection when using TEMs of this size. Use of insulating fiber washers to provide a thermal break between the fastener and fastened surfaces can compensate for the increased thermal leakage a larger-diameter fastener causes, compared to using a smaller-diameter fastener without such an insulator.
Figure 9: Illustrations of fastener hole placement inline (left) and transverse (center) with fins on a heat sink. At right, an exaggerated illustration of how heat sink flexure under clamping pressure can result in poor contact with a TEM surface.
Particularly when threaded fasteners are used, care should be exercised during assembly of a mechanically clamped TEM assembly to ensure an even application of pressure. Tightening fasteners in an uneven manner can generate extremely high forces capable of damaging a TEM at one edge, while at the same time applying insufficient pressure to minimize the thermal resistance of the interface at the other. This issue is magnified when multiple TEMs are used in a parallel configuration.
Figure 10: Tightening threaded fasteners in an uneven manner can result in uneven application of pressure across a TEM surface, or damage to the device as a result of the large lever forces that can be produced.
The production of below-ambient temperatures invites condensation of atmospheric moisture into liquid form. Corrosion of a thermoelectric device’s inner structure is a probable result if ingress of such condensation occurs, making moisture management a needful item if application longevity is desirable. Many thermoelectric devices are available with either an RTV rubber or epoxy seal applied between their ceramic endplates, in order to attend to this matter directly at the device level. RTV is considered serviceable over a relatively wide temperature range of about -60 to 200°C and is relatively inexpensive, but is not entirely impermeable and over time may permit an unacceptable level of moisture ingress depending on application conditions. In contrast, epoxy seals are considered serviceable over a somewhat narrower -40 to 130°C temperature range but are regarded as a more reliable long-term sealing agent, with a somewhat higher cost structure and lesser elasticity.
While convenient, such on-device seals can constitute a fairly significant thermal leakage path. They’re positioned across the greatest temperature differential available in the system and are typically only a few millimeters in thickness, enabling significant thermal leakage even through seal materials with relatively good thermal insulation properties. Decreases in performance due to seal materials on the order of 10% have been reported though actual results will vary, becoming more severe as the system’s operating point moves toward increasing ΔT.
The performance penalty incurred by the need for environmental sealing can potentially be reduced by moving the seal point from the TEM itself to the the heat transfer surfaces to which it is affixed. This implies application of the seal during or after assembly of the TEM to the heat transfer surfaces, with resulting considerations for manufacturing and repair/rework potential.
Figure 11: A comparison of sealed (left) and unsealed TEMS.
Using Multiple TEMs
In some applications, it may be desirable to configure multiple TEMs in either a series arrangement to achieve a higher possible ΔT, or in parallel to increase the amount of thermal energy that can be transferred.
As a practical matter, series-configured TEMs have rather limited use. TEMs are relatively inefficient heat pumps to begin with, and that inefficiency is compounded when they are arranged sequentially; each stage must not only pump all the thermal energy transferred from the cold side of the assembly, but all the electrical input energy applied to the previous stages in the chain. At per-stage ΔT values high enough to justify a series configuration, TEMs will generally exhibit a coefficient of performance significantly less than 1; stated differently, more than a watt of electrical input will be required in order to pump a watt of thermal energy through the cold side of the device. Developing a high ΔT across a TEM implies a high ΔT across thermal leakage resistances, meaning that much of the cold-side thermal transfer that’s coming at a high electrical price will have to be used to compensate for that leakage, rather than doing something potentially more useful. And since a relatively large amount of electrical energy is being put into the system, the hot side of a series-connected TEM assembly will generally need to rise significantly above ambient temperature in order to dissipate it, which effectively subtracts from the ΔT that one is struggling to produce.
Consider for example the NL4040-04BC, a 4-stage TEM module indicated as providing a maximum Q of 3.0W or ΔT of 136°C with a 50°C hot-side temperature, given roughly 31 watts of electrical input in an atmosphere of dry nitrogen. A point roughly halfway between the endpoints of the operating curve would run at roughly 1.5W and 72° of ΔT. If the environment in room to which the hot side is emptying is taken as 23°C (about 73°F) a hot-side heatsink with a thermal resistance of 0.83°C/W would be needed to keep the hot side of the TEM at 50°C. The cold side of the TEM would end up at about -22°C, requiring an insulation thermal resistance to the atmosphere of 30°C/W to hold thermal leakage from that source to the 1.5W that the TEM is capable of moving under such conditions. By loose estimation, a five-sided cubic shell 50mm (two inches) along an edge made of polyurethane foam insulation would need to be roughly a centimeter or 0.4 inches thick to provide that amount of insulation.
By rough estimation, this series-configured TEM might be used to refrigerate a region roughly the size of a small coffee cup to a temperature comparable to that of a household freezer, using about 30 watts of power to do it. Considering that common domestic freezers routinely maintain regions approaching a cubic meter in size at similar temperatures using a similar average amount of power, it’s simply not a very attractive proposition for cooling anything of size. Where devices of this of this sort do have noteworthy utility is in providing localized, controllable cooling of very small objects such as semiconductor-based sensors. Noise in such sensors is generally temperature-proportional, so for certain specialty applications where a chilled sensor is desired for noise reduction, a multi-stage TEM such as this might be used to keep a sensor at a low temperature without need of more bulky or inconvenient cooling methods.
In those limited circumstances where a series-configured TEM array is a suitable design choice, use of factory-configured devices is strongly suggested. The selection of operating parameters for each stage to obtain an optimal result is not a trivial task, and devices specifically designed and produced as series-configured arrays will generally give a better result than one assembled from discrete devices designed for single-stage use.
Figure 12: Image of an AN4040-04BC multi-stage TEM (left) and model of the same atop a heatsink (511-3M) with natural-convection thermal resistance of approximately 0.9W/°C, and a foam insulation enclosure. Such an assembly could be expected to maintain the interior area of the enclosure at a temperature similar to that of a household freezer, while consuming a similar amount of energy as such an appliance to do so.
Arranging multiple TEMs in parallel such that they share the same hot- and cold-side reservoirs is a means by which the thermal transfer capacity of a TEM can be extended. Because the absolute distance that an object expands or contracts as a result of temperature is proportional to its size, there is a practical limit to the maximum size TEM that can be produced. Consequently, there’s also a practical limit to the maximum amount of heat that can be pumped using a single device. If heat transfer capacity beyond this limit using a TEM is desired, using multiple devices in parallel is an option.
When doing so a number of practical considerations apply. First, such assemblies are generally assembled using a clamping technique, because a non-rigid bond allows for relief of thermo-mechanical stresses as component temperatures change. Second, if attempting to mount multiple TEMs between two nominally flat surfaces, achieving a uniform clamping pressure across all devices requires some care. Non-uniform TEM thicknesses and/or non-flat surfaces on mating components such as heat sinks, spacer blocks, and such can easily cause significant increases in interface thermal resistances and disappointing results, even when the dimensions involved are too small to be easily seen; maintaining relevant tolerances (TEM thickness & parallelism, heat sink mounting surface flatness) to within 0.001” (0.025mm) or less is generally recommended for best results, with a 0.0005" (0.0125mm) figure being suggested when TEMs are used in parallel.
Placement of a fastener in between each parallel-configured TEM is suggested in addition to on one on each outer edge of the array, with each preferably being the same distance from the center of the adjacent TEM. This helps to avoid irregularities in applied pressure that result from heat sink flexure. Tensioning fasteners incrementally, in a symmetrical pattern is recommended to obtain as even a clamping pressure and to avoid inadvertent TEM damage during assembly. Begin with the center-most fastener in the TEM array, and apply a modest fraction (perhaps 20%) of the final fastener torque. Proceed to torque other fasteners in the array to the same value, moving outward in an alternating, symmetric fashion. When all fasteners have been tensioned in this way, repeat the pattern applying further torque to each fastener.
Figure 13: Non-uniformity of TEM thickness results in uneven pressure across the TEMs’ mounting surfaces when assembled, and can result in device damage or sub-optimal system performance.