Control Algorithm of Modern Switching Regulators

In the past few years, the pitch on digital control of switching regulators has been rising steadily. In a typical analog implementation, a pulse-width-modulated (PWM) switching regulator is built around a modulator composed of a comparator, having a periodic piecewise-linear (triangle or sawtoothed) modulation waveform at one input and the error signal at the other input. As the quasi-stationary error signal falls between the minimum and the maximum of the modulation waveform, the comparator output produces a square wave at the heart of this modulation scheme. This system is “analog” simply because it's governed by the analog modulation waveform. The questions is: Are there viable instances where using true digital architectures in switching regulators is preferable to analog?

When people talk about digital control, they generally refer to one of three architectures:

• An analog engine, as described above, but equipped with digital peripheral functions such as serial communications (I²C bus and SMBus). In CPU voltage regulator modules (VRMs), the digital inputs are called voltage identification codes or VIDs, which are essentially a digital means to vary on demand an otherwise constant reference. I'll call this a digitally controlled analog system.

• A microcontroller-based digital architecture. This is a useful architecture in terms of flexibility, especially in applications where programmability, as well as current and voltage profiling, is required. As conventional digital algorithms are sequential in nature, requiring several clock cycles to execute an instruction, they are inherently slow and thus aren't useful in applications requiring fast response. I'll refer to this as a microcontroller-based control architecture.

• A non-sequential machine, hard-wired logic implementation that can produce a fast response comparable to an analog system. I call this a true digital control architecture. It follows that the challenge to the analog switching regulator dominance in fast-response applications may only come from the true digital implementation described here.

Although true digital control is an interesting technology, I have yet to hear a compelling case for it. Currently, this technology remains at the periphery of the action. But is there a relevant place for true digital control, one in which such control is not just more convenient but is fundamentally superior in performance?

To answer this question, examine the system you want to regulate. If the system is truly linear, namely continuous and invariant in its mode of operation, or smooth, then analog is the way to go. This is true in the case of a desktop CPU voltage regulator whose output voltage must be controlled continuously by the same algorithm from no load to full load.

However, if the system is non-smooth — namely discontinuous and variable in its mode of operation — then digital may be the way to go. For example, digital could be used in the case of a cell phone voltage regulator where, due to the necessity to save power at light loads, a mode change is required, typically from a PWM algorithm to pulse frequency modulation (PFM). PFM is a mode in which the frequency adjusts with the load, thereby yielding lower frequencies and hence lower switching losses at lighter loads. Such mode change in an analog system requires an abrupt commutation from one control loop (for example, PWM) to the other (PFM), typically at the time the load is changing. This type of algorithm discontinuity invariably leads to some degree of temporary loss of output regulation.

By contrast, a digital control is inherently equipped to handle discontinuities and thus handles mode changes within a single control algorithm.

In conclusion, I believe digital control may bring relief to its analog counterpart in non-smooth systems. In such systems, digital control may prevent risk of loss of regulation and save additional overhead in bill of materials that would be required to mitigate the effect of discontinuities in analog implementations.

Reno Rossetti has more than 25 years of experience in design, applications and marketing in the analog semiconductor industry. Reno's career has evolved around power ICs and discrete semiconductors. He worked at ST in spindle motor and voice coil driver design, and later headed up National's power management design group. At Fairchild, Reno covered several positions in new product development and applications. He is currently director of corporate strategy for the computing and ultraportable segments. Reno has the BSEE degree from Politecnico di Torino in Italy and an MBA degree from the Bocconi University of Milan.