Experts Find There is More Analog in Digital Power Than Previously Thought

Last week at DesignCon there was talk of 100GigaBit SerDes: 4 channels of 25GigiBit. We all know that bits and frames fly over the wire, and at first glance these are digital communication systems. Yet Power Integrity problems involve Power Distribution Network (PDN) impedance over a large frequency range (DC-10Ghz) due to the board layout, packaging, and die. This is a vast land of 2D/3D Electromagnetic modeling, reflections, discontinuities, reference plane changes, and under the right conditions, signals radiating from a design like a radio transmitter.

Remember near and far fields? In one presentation we were shown an old-fashioned dipole antenna and taught how to design a ground plane system that made a great dipole antenna. A demonstration totally blocked out a radio station on a small hand held transistor radio because the power cord was an antenna.

Digital indeed!

I take great comfort that the world is analog at its core and this old dog is not obsolete. But clearly, when I write C code for a Cortex M4 Board Controller, conceptualizing the problem domain and solution is, well, Digital. It remains digital to me as long as the bypass capacitors maintain their integrity, the power supply remains in tolerance, and the temperature is stable. It turns out the digital view of the world is just a special constrained case of analog, and it is more so today than ever. It is no accident that Planet Analog was advertising at DesignCon.

In my own backyard I find a similar dilemma. The problem manifests itself in the term “Digital Power.” Can power really be digital, or is this another case where the world is really Analog? Perhaps “Digital Power,” is just poor fiction.

In the good old days, power conversion was either an analog feedback loop around a pass transistor (LDO), or an analog feedback loop around switching devices with filters (Buck Converter). Then someone got the bright idea to control a switching converter reference point with DACs, measure output voltage and current with ADCs, throw in some comparators for faults, and add a serial communication bus.

Then some crazy Digital Signal Processing (DSP) guys said, why not replace the feedback compensation with a math processor. The result is a market filled with different combinations of analog and digital features. We called it all “Digital Power” and have no idea what it means when we say it.

The problem is when we use the term “Digital Power,” nobody ever agrees about what it refers to. To some people it is PMBus, and to others it is Digital Compensation. VRM guys might think it is a two wire Automatic Voltage Scaling Bus (AVS). Heck, it could even chant “Digital Power” like “Power to the People”. Digital rules the world! Down with Analog!

The main confusion is whether “Digital Power” means a switching power conversion system with analog compensation and PMBus, or a switching power conversion system with digital compensation and PMBus. And what about the management chips full of DACs, ADCs, and servos? As soon as I say I have a “Digital Power” device, depending on which interpretation is in play, vastly different mental images appear and I spend so much time explaining what I mean, that I might as well never have said “Digital Power.”

To make things worse, there are products like the LTC3880 that have two loops, one with an inner analog control loop/compensator and another outer digital servo loop! At the end of the day though, what matters are the bandwidth, phase margin, output impedance, the 0.5% accuracy, and the fact that the behavior is linear, predictable, and robust. The digital aspect is just a means, not an end.

At Linear Technology we wrestled with this for a while and finally decided that it would be better to describe things by their function. Like, “PMBus Manager” and “PMBus Controller.” Managers are devices like the LTC2978  that wrap around any power supply and provide sequencing, margining, supervision, telemetry, and fault management.

Controllers are devices like the LTC3880  that buck down an input voltage to a lower output voltage. Other attributes like voltage mode or current mode and number of channels/phases, should be treated as properties or features of the product. Control loops will be described as analog or digital/servo according to the method of control.

So if you find us avoiding the term “Digital Power,” and instead using terms like “Power System Management,” “Buck Converter,” or “Power Supply Manager,” we are just trying to avoid confusion. Besides, we know the world is really analog and digital is just a convenient fiction. :)


Michael Jones