Early in the process of designing a high-power, high-value product such as a telecoms server board or an industrial computer, the development team will decide whether to implement the intermediate power stage and Point-of-Load (PoL) converters with complete, fully integrated modules, or with discrete power components instead.
When system architects consider this decision, it can be tempting to take a matrix approach: to list the advantages and disadvantages of each, and weigh them up in the light of the particular requirements of the end product. So, for example, the benefits of using a power module include:
• Fast time to market, since a module provides a ready-made circuit which is supplied fully tested and approved. A module reduces the burden of EMC compliance on the user.
• Small board footprint and low component count
• Easy for engineers with limited analogue or power design experience to implement in product designs
Among the drawbacks of a power module, designers might be concerned about their limited ability to second-source the device, and its potentially higher unit cost.
For a discrete circuit implementation, these benefits and drawbacks would be reversed: it would be larger and more complex and would use many more components, but those components would normally be fairly readily replaceable, and the Bill-of-Materials (BoM) cost would normally be lower than that of the module.
If the overriding design consideration is BoM cost, it is likely, with this decision method, that the development team will opt for a discrete circuit. If, on the other hand, time to market is the most important issue, demanding a rapid design process and minimal testing and validation, or if power density is a critical factor, the circumstances favour the choice of a module. If the design team has little or no power or analogue expertise, a module will also be the preferred choice.
This is, of course, a rather stark characterisation of the system architect’s decision-making, and in reality the thought process behind it will be much more complex and nuanced than this simple description implies. Nevertheless, development teams commonly make the assumption that power modules and discrete power-converter circuits which perform the same function are therefore equivalent, and can be compared on a like-for-like basis.
In this author’s experience, this can be a flawed assumption. This article explains why.
Why module-based designs perform differently
On first consideration, it might seem odd to suggest that a power module-based design would perform differently from a discrete circuit working with the same input and output voltage and current specifications. To take a random example, Intersil’s ISL8273M is an 80A digital power module which operates from a supply rail of 4.5V to 14V, and which provides a single-channel, dual-phase output at a voltage which is programmable in the range 0.6V to 2.5V. Inside its 18mm x 23mm x 7.5mm package will be found a digital power-controller IC, MOSFETs, LDOs, inductors and various other power components – the same component types with which an equivalent discrete power circuit would be implemented.
Because of the way the module is packaged and assembled, its board footprint is certainly a great deal smaller than that of the equivalent discrete circuit. But this is not the only difference. In fact, there are three ways in which a module in many cases offers superior performance to a functionally equivalent discrete circuit.
1) Layout optimisation
Reputable suppliers of power modules, such as Intersil, Vicor, Exar, ON Semiconductor and Fairchild, have valuable brands to protect. This means that they take great pains to provide users of their modules with every assistance to ensure that they work to their highest potential in all conditions.
One way in which they do this is by providing layout guidelines which give the user a blueprint for optimising the thermal and electrical performance of the entire power circuit, including the small number of external components supporting the module. This optimal layout will have been developed and refined after exhaustive testing across the range of operating temperatures and other conditions specified for the module, as shown in Figure 1.
Because the manufacturer’s brand stands behind the module, it is important for it to commit the necessary engineering time and resources to this optimisation. By contrast, the economics of product development at OEMs dictate that the same level of layout optimisation is normally out of the question. The thermal and board-level design is typically done on a ‘best effort’ basis, to achieve the best performance in the limited development time available.
In most cases, this best effort will fall short of the absolute optimisation that module manufacturers perform. The thermal performance and layout efficiency of a module-based system can therefore typically be assumed to be better than that of a discrete circuit.
2) System reliability
Much the same argument applies to the reliability of a power module when used within its specified operating conditions.
The reliability argument in favour of modules over discrete power circuits is usually illustrated in terms of component numbers: put simply, this argument states that a system’s Failure In Time (FIT) rate increases in a more or less linear fashion with each increase in the number of components in the system. In other words, a circuit containing 20 components is likely to have a FIT rate around 20 times higher than that of a circuit containing a single component.
Of course, a module is not itself a single ‘component’: it is a single package containing many components. Nevertheless, experience shows that the real-world reliability of power modules is far superior to that of comparable discrete circuits. This is because the construction of a module is, in various ways, less prone to failure than a set of discrete board-mounted components. For instance, the module’s footprint is smaller than that of the equivalent discrete circuit, so it is less affected by any warping of the board. Also, the temperature inside a module is more evenly distributed, because it is smaller and the moulding compound aids heat distribution. With fewer, less intense hot spots, the devices inside the module run cooler and more efficiently and are less likely to fail.
A module also has fewer solder joints: the dies of active components are directly soldered to the base board of the module, but the other connections are made by bonding, as in an IC, as shown in Figure 2. This means that there are far fewer solder joints and bonds than in a discrete design. In addition, bonds are less susceptible to cracking, bending and other failure mechanisms than the solder joints used with discrete surface-mount components.
But the reliability advantage of the module goes further than this: again, module manufacturers commit enormous time and resources to testing and validating their products, as shown in Figure 3. This testing can include Highly Accelerated Lifetime Testing (HALT), which overstresses the part in order to arrive at a calculated Mean Time Before Failure (MTBF) value for the module.
Reliability reports provided on the manufacturer’s website also give an indication of the product’s FIT rate.
What this means is that the power-system designer can benefit from fully tested, fully documented reliability data about the entire power circuit, across all specified operating conditions. By contrast, any reliability data about the components used in a discrete circuit apply only to each of these components individually, and not to the entire circuit.
In the case of a module, these system-level reliability data are based on exhaustive and lengthy tests; again, few OEMs will have the engineering resources available to perform such exhaustive testing on an in-house power-supply design. For these reasons, then, the designer can have greater confidence in the reliability of a module-based power-system design than in that of a discrete circuit; and the real-world reliability of the module-based production units is more likely to match or exceed their predicted performance than production units of a discrete circuit.
There is a further advantage for the user of a module: he or she has the reliability data available from the beginning of the design process. By contrast, testing to establish the reliability of a discrete circuit can only take place after the circuit has been designed and built in prototype form. If at this stage it is found to be too unreliable, and the design requires modification, there is a severe risk of delay to the introduction of the product. With a module, there is no such risk.
3) Predictability of performance
The third way in which a module’s performance can be expected to be superior to that of its discrete equivalent is in the predictability of its performance. For the central component of a power-conversion circuit, the power regulator or power controller IC, performance is normally well documented in the datasheet. But this documentation only applies to the IC.
For a module, this documentation applies to the entire power-conversion circuit. This means that the module user can implement a system design in full confidence that the performance of the power system’s electrical parameters, such as line and load regulation and EMC, will be exactly as predicted by the product documentation.
For a discrete circuit, the performance of the system, as opposed to that of a single controller or regulator IC, is not at all supported by a manufacturer’s documentation. As above, an OEM’s testing and validation of a discrete system’s performance will rarely be as exhaustive or comprehensive as that of a module manufacturer. Defects or weaknesses in the performance of a discrete circuit in a given operating condition might therefore elude discovery during the design process. This creates a risk that the real-world performance of the discrete circuit will be worse than that of its module equivalent.
Avoiding the trap of like-for-like comparisons
This article argues, then, that the decision over whether to use a module or discrete components for the implementation of a power-conversion circuit should not be based on the assumption that functional equivalence is the same as performance equivalence.
According to the arguments presented above, it is certain that in some cases a module will perform better – both electrically and thermally, and over a more predictable lifetime – than its discrete equivalent. The assumption that exact like-for-like comparisons can be made between a module and its discrete equivalent is therefore flawed.