Power Supply Design for maximum Performance

Designing for energy performance (typically measured in terms of efficiency metrics) implies a focus on optimizing for the performance of power supply energy consumption, which translates to emphasizing operational expenditures (OPEX) or, essentially, the cost of energy.

If a power solution is optimized for its form-factor performance, this may be in opposition to maximal conversion efficiency, which effectively means optimizing a design for capital expenditure (CAPEX) or essentially focusing more on upfront cost savings instead of the amortized cost savings that occur with reduced OPEX. This distinction can be critical in applications where the power OPEX dominates the total cost of ownership (TCO), such as in the large scale data center world.

For untethered applications, power OPEX can be articulated in terms of fuel, range, and/or battery life. Typically, these limited energy sources will act as the controlling factors in maximizing system performance.

Therefore, it is imperative for engineers to understand the (sometimes very complex) relationships between supply, load, and the operating environment before they can begin to articulate what performance factor(s) shall be the focus of optimization. For power solutions, most design parameters ultimately converge on a design for thermal performance, keeping critical components (semiconductor junctions, package surface temperatures, printed circuit board, or PCB temperatures) below critical temperature thresholds under worst-case operating conditions, such as maximum input voltage, full load, and high ambient temperature.

Addressing Uptime Performance

If output voltage regulation and accuracy are the most critical, then optimizing power supply control loop performance (feedback loop stability and load transient response) may take precedence to ensure the power delivery does not wind up in an unstable/unpredictable state after abrupt changes in load or during supply voltage dips and surges.

When operational reliability is the top priority, it typically implies a mission-critical task in which the key performance indicator is the uptime of the application/system itself. In this scenario, the system requirements may even call for the sacrifice of the power supply and other equipment for the sake of keeping the application alive for as long as possible, even if the operating conditions are out-of-specification. This is a completely different approach to designing power supplies with built-in shut-down protection if there is a short-term overload, overcurrent, or over-temperature condition.

While not typically recognized as key bottlenecks in application performance, power and thermal metrics are often the primary performance limiting factors due to basic physics. The reason behind the limitation may be a maximum junction temperature of a power semiconductor device or the maximum current of the line cord or a power inductor, but, in one way or another, performance is eventually throttled by power/thermal limitations.

There are times when system performance is derated because of the need to maintain an overall thermal envelope or thermal partitions/zones: For example, a processor may be able to provide additional millions of instructions per second (MIPS), or a radio may have additional headroom to further amplify an RF signal, but the system lacks sufficient thermal management techniques to allow for the added dissipated power.

Emphasizing Energy Performance

Power is often taken for granted, not only regarding an underappreciation for the complexity and specialty needs of power solutions, but also in terms of availability. As alluded to earlier in the commentary on thermal bottlenecks, it is not uncommon to see systems designed with a significant gap between the peak demand from the load and what the power source can provide to save costs or squeeze the power supply into a smaller space. When loop control and transient design challenges (for the context of this blog) are neglected, a power gap can also occur if the power subsystem design analysis was conducted at a too-low margin of safety and did not have sufficient consideration for all loads sourced by a common rail, or even the aggregation of upstream rails into larger power solutions.

Most power subsystems involve multiple levels of voltage conversion from offline (ac) to mid-bus voltages (typically 48/24/12 Vdc) to low voltage for ASICs and other logic circuits (typically ≤5 Vdc). Typically, more attention is paid to the efficiency of the power conversion solutions with lower level voltage rails, because the load current tends to increase with decreasing bus voltage; in such cases, dissipated losses become more dominant and critical to the overall system thermal performance.

Even with this higher attention at the load end, it can be easy to overlook the impacts of upstream power conversion solutions. Therefore, it is vital to develop an interactive model of a system power budget that considers the load vs. efficiency curves of all power supplies and the aggregate effects of transient performance, from end load all the way upstream to the offline source.

Is Power Supply Performance Disaggregated from System Performance?

Of course not! But although stated emphatically, there may be some misperceptions to the contrary. Everything in the world of electronics requires power—and therefore power solutions—and there is a direct relationship between the performance of the power supply and the success of the system/application as a whole.

Typically, this relationship is perceived in a highly oversimplified way (supply turns on, and the system is powered up), which does not consider the stability of the power supply when it interacts with the load.

For instance, if the load demands greater transient performance than the power supply was designed to give, this can cause instability of the power supply control loop with any number of adverse effects, such as poor regulation, inability to startup, undesired tripping of protection circuits, excessive electromagnetic emissions, or EMC issues.

A power solution’s viability is just as dependent on the environmental scenario as the rest of the system components. It is not uncommon to derate a power solution’s fully rated output at lower supply voltages, which is ultimately limited by thermal bottlenecks as described above.

In addition, we have already discussed how having more available power does not necessarily imply increased usage by the load if there is a limit to the amount of dissipated power the system can practically mitigate. Operating at higher elevations with lower atmospheric pressure means performance must be further derated, because one kft or 300 m is roughly equivalent to an extra 1 °C of ambient operating temperature at sea level. In addition, the isolation grade needs to be reinforced for operation at high altitudes, as flash-over is easier at low atmospheric pressures. For this reason, RECOM adds an operating altitude specification to its AC/DC power supply specifications.

In Summary

All that being said, there are deployment scenarios and power architectures intentionally designed to try to disaggregate power supply performance from system performance. For the most part, this is in reference to applications with redundancy that have greater than a 1:1 ratio of power source budget to system power budget. Redundant power supplies serving a common load bus tend to current-share (rule of thumb is even sharing within 10 % of each other), meaning that a supply is operating well below its maximum rated output current.

The most fundamental form of redundancy is an N+1 configuration in which two (typically identical) power supplies provide shared power to a system, even though an individual unit is capable of sourcing the full load. With many typical layers of margin built into system power budgets, this implies these power supplies will run at a maximum of 30-40% of their maximum rated output current, even when the system is demanding its absolute maximum power draw. If both the power supplies and system in this scenario were comprehensively designed and qualified for the full system power, the demonstrated life performance of the power supply would be very different from that of the system, because the system components are effectively exposed to thermal stresses twice that of the power supplies.

Another brief example of disaggregation between power solutions and system/load demand is in the load shedding/sharing scenario; it is not always pragmatic to size the power solutions based on a system power budget created from the sum of maxima of worst-case load profiles. Rarely is there a case in which all loads see their maximum power draw concurrently, leading to many unnecessary overdesigns.

However, these overdesigns can sometimes dictate a larger, more expensive and/or less efficient power solution than the application requires. If major system loads are known to be in antiphase (compute and memory are a classic example of major system loads that tend to have the peaks/valleys of power waveforms 180° out of phase with each other), then a smaller power solution can be utilized with the use of intelligent power management (IPM) techniques.

Some power supplies are even intentionally designed for short-term ride though of an overcurrent condition without turning off or instantiating any overcurrent protection (OCP). One example of this is a system supporting multiple Power-over-Ethernet (PoE) ports of variable power levels, which may demand more short-term power than the supplies can provide (like 120 % of rated load for <200 ms), while negotiating individual PoE port power levels. In this case, the power supply is designed to ride through such events without tripping OCP, while still offering short-circuit-protection (SCP) for longer term overcurrent events. The presence of a digital control core, such as the RACM1200-V power supply from RECOM, allows easy programming of the response of the power supply to such events.