Power Modules are Catching up with Moore’s Law

Moore’s Law, or more correctly his prediction, was that the number of transistors in a complex IC would approximately double every year. Gordon Moore made this statement in a popular magazine in 1965 and was only looking ahead ten years at the time.

His prediction was that, from around 50 components in an IC in that year, by 1975 there could be 65,000 on a 6 x 6mm die. His prediction came close, with the Intel 8086 released in 1976 featuring 29,000 transistors using a process scale of 3.2µm. Moore then revised his estimate to doubling every two years. Today, although the rate of increase in complexity has slowed, a scale of 2nm is now targeted by TSMC for 2025 [1], an amazing reduction in die feature size by a factor of 1600.

As of 2022, the highest transistor count in a commercially available microprocessor is 114 billion, in Apple’s ARM-based dual-die M1 Ultra system, fabricated in a 5nm process. This ‘System on a Chip’ consists of two dies, each 420mm² in area. Process innovations such as 3D transistors and ‘die stacking’ have kept the metric of processing power per mm² footprint on a similar exponential trajectory.

In principle, conversion topologies used in PoL converters have changed little over several decades. Buck, boost and buck-boost circuits are used, which still use a semiconductor switch, diode or synchronous rectifier, inductor and capacitor, with pulse width or frequency modulation to achieve regulation. Switches have evolved to exhibit lower static and dynamic losses and operate at higher frequency, and inductor core materials have improved incrementally for lower loss, as have capacitors. As a result, efficiency has increased, and this in turn has enabled better power density – PoL converters can be smaller for the same load and temperature rise. The benefit has not followed Moore’s Law though, and power converters remain a major proportion of board space used.

Part of the problem with the size of the PoL converter is that discrete components have been traditionally used on a substrate with little integration. This follows on from the simple fact that they are power converters and inevitably dissipate heat, so large components have been necessary to keep temperature rises low. Also, the inductor and its core are difficult to fabricate as other than a bulky discrete component, and ferrite core technology has only improved in small steps over the decades. This means that a minimum number of coil turns is necessary, which must also be thick enough to withstand the higher currents without overheating. This makes for a bulky component. The number of turns required does decrease as switching frequency rises for the same core flux density and output ripple current, but then core and PoL converter switching losses increase, requiring a larger PoL anyway to dissipate the heat. This all means that, unlike processors, scaling down PoL converters in size tends to directly introduce thermal problems, limiting overall gains.

Point-of-load converter modules do not have standard input ranges, and outputs needed could range from around 0.5V to 3.3V or possibly to 5V for legacy designs. Current ratings could also vary from sub-one amp to tens of amps.

The input to a power module may be derived from a regulated bus at 12V, 5V, 3.3V or sometimes an intermediate value such as 9V. Increasingly, for battery-powered devices the input could be from a lithium-ion cell at around 3.7V but could be over 4V when charging and 3V or less when heavily discharged. Equally, the input may come from an unregulated source and vary over a 2:1 range or more.

To meet a wide range of applications, RECOM now offers parts in their RP and RB series with inputs down to 0.85V for a boost converter and up to 65V for a buck converter. All parts have adjustable outputs and cover applications needing outputs of anything from 0.6V to 35V. Current ratings are from 0.5A to 15A. Figure 5 summarizes the buck converter combinations available.


Figure 6 similarly shows the range of boost power modules available from RECOM with up to 10A rating. The RBB5-1.5 and RBB10 parts are buck-boost types where the output can be higher or lower than the input, which is useful for maintaining regulated power from a battery source as it discharges.

Boost converters offered up to 1.5A include output overvoltage protection, and all include a true load disconnect function when disabled, which is necessary as there is usually a direct DC path through a boost converter when not switching. All buck and boost converters additionally have over-temperature and output overcurrent protection.

So, have power modules caught up with Moore’s Law? Let’s take a couple of examples comparing PoL converters with 24V input and 3.3V output at 4A. Just four years ago, in 2018, a part was advertised as ‘leading the advancement of power conversion’. It had a footprint of 33mm x 13.6mm, it was 8.8mm high and had a volume of 3950mm³. Today the RECOM RPX-4.0 has a 5mm x 5.5mm footprint, is 4.1mm high and has a volume of 113mm3, with a 35-fold increase in power density. Interestingly, the efficiency of the two parts is about the same, the improvement all down to higher switching frequency with smaller components and 3D power packaging^® techniques.

Another example might be converting 5V to 1.8V at 2A. Older through-hole parts can still be seen with a footprint of 14mm x 7.5mm and a height of 10.1mm. Compare this with the RECOM RPZ-2.0 at 3mm x 3mm and 1.1mm high, and an increase in power density by an impressive 107 times (Figure 7). Footprint has decreased by a factor of nearly 12, the RECOM part only occupies one side of the motherboard, and the inconvenience of possible manual placement and through-hole soldering is avoided.

With older open-frame DIP-style technology PoL converters, it was often evident to users that the same components could be fitted on their motherboard, making the value of a bought-in module questionable. Even space-saving SIP parts could easily be replicated by the user to save costs, using the same component styles and assembly technology that they were already using on their motherboards. However, power modules using 3D power packaging simply cannot be replicated economically in motherboard fabrication technology, and the modules can now be regarded as components such as resistors or capacitors that a user would never think of fabricating themselves. The size of power modules now enables them to be fitted easily around or under other components, and the highly automated manufacturing techniques now used drive costs down and reliability up.

Power modules now offer power density performance that far exceeds what is possible with discrete solutions. Gains in recent years have rivalled those in IC transistor density, in line with Moore’s Law. With manufacturing automation, costs are increasingly competitive, and when purchasing, stocking, placement and test overheads are factored in, a module solution easily comes out on top. The extended range available from RECOM covers all common power and voltage levels found in applications from hand-held devices through to server and telecom boards.

[1] https://pr.tsmc.com/english/news/2939
[2] https://recom-power.com