Why select a digital power converter?

Faced with ever increasing choices between analog and digital power converters, many engineers remain uncertain about the relative merits of digital control techniques. 'Digital control' in this context refers to closing the power converter's internal feedback loop, rather than the subsidiary external control and monitoring functions that are increasingly digital by default.

This change in converter implementation is the key element in any discussion regarding digital power, but what benefits accompany moving from tried-and trusted analog loop control to a digital version? And what impact, if any, is there for designers wishing to use digital power converter modules in place of existing analog designs?

Tests compare converter performance
As always, it's necessary to compare competing designs for parameters of interest. In this case, these include electrical performance, conversion efficiency, parts count, power density, reliability and cost. Arriving at truly representative results requires a comparison between two converters that—apart from their inner control loop implementations—are as similar as possible. For this reason, Ericsson chose a traditional synchronous buck converter from its range to provide baseline test data, making modifications to the converter's core design to replace its analog control loop with a digital equivalent.

The baseline PMH8918L supplies up to 18A for non-isolated point-of-load (POL) applications. With a nominal 12V input and programmable output voltage, the device lies within a highly popular voltage and current range that should provide useful data for a wide range of customers. The equivalent digital design takes advantage of the ZL2005 chip from Zilker Labs, Inc. To ensure the most representative test platform, most of the power train components are common between the two designs.

The PMH8918L has the following basic specifications:

? Output Current: 18A

? Topology: Synchronous buck downconverter

? Control method: analog pulse-width modulation (PWM)

? Input voltage: 10.8-13.2V

? Output voltage: 1.2-5.5V (user programmable)

? Switching frequency: 320kHz

? Dimensions: 38.1mm x 22.1mm x 9mm

The ZL2005-based digital control implementation occupies the same PCB area as the PMH8918L and switches at 333kHz, which is very similar to the analog regulator's 320kHz. A first set of tests uses the same Renesas high- and low-side output MOSFETs that appear in the PMH8918L for both converters ('digital Renesas'). But to take advantage of Zilker Lab's recommendations for optimizing the digital converter's output switching dead-time, a second set of results uses Infineon MOSFETs ('digital Infineon'). These devices have a higher gate resistance (1.2Ω rather than 0.5Ω) but are otherwise very similar in terms of key parameters, such as drain-to-source on-resistance and switching losses.

The respective FET part numbers are:

The tests compare the respective converter configurations using a 12V input voltage in a room temperature environment. A programming resistor sets the output voltage to 3.3V or 1.5V, and the converter's sense pin connects to its output pin to configure local voltage sensing. While the conclusions should apply over a broad range of operating currents and power modules, tests to date were performed on a small group of samples of this model of regulator.

Measurement results
Compared with the analog approach at the 3.3V level, the Renesas digital implementation shows an efficiency increase of around 1 percent from 2-10A. It reaches the 90 percent point at 2A rather than 3A, peaks at around 94percent for an 8A output, and maintains an advantage up to the 92 percent point at 16A. It then performs no better than the analog converter up to full load at 18A.

The digital Infineon version performs better across the entire range, maintaining a clear advantage from Figure 1). These efficiency improvements contribute towards power dissipation reductions of 1W or more from 5-18A. The digital converter also saves power by eliminating the housekeeping and protection circuitry that the analog DC/DC circuit requires.

Figure 1: The digital converter shows clear efficiency gains over its analog counterpart.

At the more difficult 12V to1.5V conversion step, the analog implementation struggles to achieve 87 percent, which it maintains from about 7-11A. By comparison, the digital Infineon converter reaches 87 percent at around 3A, achieves 90 percent from 5-10A, and at 86 percent is still 1 percent more efficient at full load. The resulting power dissipation reduction is approximately 1W from around 2-18A.

The steady-state voltage regulation performance for the analog converter and each of the digital versions is essentially the same for both output levels.

Dynamic performance
Output ripple and noise and output load transient response tests were performed for a 3.3V output voltage level. The filter used for the ripple and noise measurement consisted of a 0.1?F ceramic and a 10?F tantalum capacitor in parallel, as the PMH8918L's datasheet defines. The Infineon digital converter exhibited slightly higher ripple and noise than its analog counterpart, mainly due to component tolerances in the external capacitors. There is also a minor variation that the small difference in switching frequency creates, but at around 20-30mV peak-to-peak, the ripple and noise performances are essentially the same for all practical purposes.

The dynamic load for the transient response measurements consisted of a step change from 18A to 9A and then back to 18A. The analog solution provided a traditional smooth voltage response to the dynamic load current change, with amplitude peaks of approximately ±70mV. The digital solution, which was programmed to work in non-linear response mode, shows similar peak amplitude at low-to-high load transitions and somewhat higher peak amplitude at high-to-low load transitions. Due to the NLR mode operation, the peak is distributed over time, generating a burst of peaks smaller than it would have been with NLR turned off. Because these NLR settings are not optimized, it should be possible to improve the dynamic response waveform. Even so, the amplitude of the voltage response is similar to that of the analog regulator.

Implementation, reliability
To create the most even comparison, these calculations contrast the benchmark analog regulator with the digital control version that uses the same Renesas FETs. Cost estimates are general due to uncertain trends in component costing, but there is sufficient data to project relative cost differences between the two approaches.

Component count and packaging—The component count for the digital regulator is 21 vs. 58 for the analog regulator, which equates to a 64 percent reduction. This reduction will drive improvements in cost, packaging size and reliability. Even though the PCBs of the two regulators have the same area, there is a significant difference in packaging density due to the digital solution's lower component count (Figure 2).

Figure 2: Using 64 percent less components, the digital converter saves space.

Clearly, the digital POL regulator layout is not optimized, and two approaches could greatly improve a production version. Reducing the PCB area by 40-50 percent while maintaining the 18A current rating would significantly improve packaging density; alternatively, it should be possible to approximately double the power-handling capacity within the existing outline.

Cost estimates—In terms of bill-of-materials cost, a 10-pin version of a digital regulator (i.e. equivalent to the analog version) should definitely be less than the present PMH-series design, due to the reduction in parts count. While slightly more expensive, a 13-pin digital version with a communication interface should also be cheaper than the analog implementation. Cost savings should also come from mounting fewer parts during manufacture.

Reliability estimates—Ericsson performs extensive failure rate analysis and reliability predictions for all of its products using the methodology that appears in Telecordia SR332, issue 1, black box technique. Mean-time-between-failure (MTBF) predictions are made under full output power at an ambient operating temperature of 40°C. The predicted reliability for the PMH8918L analog and Zilker Labs ZL2005 digital approach are 3.87 million hours for analog and 4.31 million hours for digital.

The reduction in component count makes the digital version more reliable even with the addition of complex components, such as memory in the digital control chip. An 18A digital version built using the same PCB area as the existing analog module would lower the circuit's operating temperature, further increasing MTBF.

Conclusions
The laboratory measurements and calculations provide four main conclusions:

1. The electrical performance, including efficiency, of the digitally controlled converter is equal to or better than the analog version. Additional work should optimize the dynamic load response of the digital design.

2. The digital solution results in over 60 percent reduction in parts count, significantly reducing the materials and assembly costs of the converter.

3. The parts count reduction reduces PCB real estate, allowing a converter size reduction or a power output increase within the present format. Either way, digital control increases power density.

4. Parts count reductions significantly increase the predicted reliability.

These benefits require no additional effort from the OEM customer. The digital regulator module may be used interchangeably with the analog version, and requires no special interface or design accommodation. It's also worth noting that the digital heart of the converter naturally lends itself to interfacing with external control and monitoring circuitry, so in this sense, much of the interfacing circuitry necessary within an analog converter comes "for free" in a digital version.

This article is contributed by Ericsson Power Modules.