Anablog

When PR hotshot Matthew Quint called to tell me about a new startup called CHiL Semiconductor in the motherboard power field I was skeptical. After all, once the Taiwanese companies ran the costs down there was little money to be made in VID regulators for Pentium and AMD chips. Heck, Analog Devices just sold their division that did these chips to ON Semiconductor. I thought it was a jellybean market with jellybean profits. Well I was wrong. It turns out there are two great markets for high performance PC regulators. Blade servers and other servers need very efficient power that can squeeze every percent out of efficiency. CHiL Semi Marketing VP Larry Spaziani told me that a watt saved in the processor means about 7 watts in the power bill coming into the building. The other high-performance application is in gamer motherboards. The gamers want to jack up the VID voltages in order to over-clock the chips. Since the CHiL regulators handle transient responses so well, the gamers can do this without blowing up the CPU or more insidiously, causing voltage stresses that will shorten the life of the CPU.

The server market needs efficient power but it also demands that the chip measure and report back on voltage, current, and other parameters so the maintenance team can catch problems before a catastrophic failure. Since the CHiL chips are done in standard CMOS from a public fab they can implement all the communications and monitoring that is needed. They have also developed IP blocks that use the cheap standard CMOS to make decent analog circuit blocks. This way they can offer a cost-competitive part that handles the transient loads of the PC CPU.

One thing I really liked about the CHiL presentation was that it did not just crow about being digital power and then sit back and expect an old analog curmudgeon like me to be impressed. I got really sick of companies writing articles that claimed that everything was going to be digital power and the only reason they weren’t selling chips was that us analog guys were such dinosaurs we could not do a digital design. After attending Darnell’s Digital Power Forum last year I got my thinking straight on digital power. Here is the deal— to us users it really does not matter whether there is a digital loop or an analog loop. If you need really low power consumption and instant-on you probably will stick with analog power chips. If you are doing a wind turbine and already have a few DSPs you can just put in the libraries that will do pure digital DSP-based power. If you are in between you will probably be able to look at state-machine based digital power that uses much less current than the DSP solutions but has the same benefits as digital power. The reason blade servers are a good candidate is that they are high-power systems, hundreds of watts, so a few milliampers to run the state machine is no big deal. The other angle is that servers use 4 or 6 or soon 8-phase power. This makes the output caps small but, better yet, allows efficiency improvements since the controller can turn off phases as the load goes down. This way CHiL get better than 90% efficiency over a very broad range of loads. Since there is so much figuring and cogitating going on to do all the measurement and control you need a bunch of digital on the chip anyway. Since the IC company is already paying for an expensive fine-line digital mask set, it is no big deal to drop the other shoe and include a digital PWM and digital filters and such to make the control loop.

For people that need simple power, like one to seven rails and not multiphase, but still need a lot of communications there are solutions like Summit Micro, that use an analog loop for microampere quiescent currents and then hang a bunch of digital around that loop to report, control and monitor the power. And for really simple power nothing will beat a plain-Jane analog chip. High voltage and off-line power chips will be made in bipolar not CMOS, so they are likely to stay analog. Anyone that has put a PAL or a microcontroller on a board knows that there are a ton of documentation and configuration management issues. A whole set of rev-control nightmares will be created, but if you are trying to differentiate your product, and a digitally-configured chip does that, the configuration and development issues are a small price to pay. Another interesting benefit is that you can “hide” the IP of your compensation and control in a digital chip, but the discrete components you use for a simple regulator are there for all to see and copy.

The upside to digital power seems evident in systems that can afford the quiescent current and have a lot of rails to generate and control, especially with multi-phase architectures. The downsides of a programmable part are no greater than a PAL, but then again, the programmability may save your butt when you can change the loop compensation without changing the BOM or spinning the circuit board. CHiL claims that since they can do non-linear response they can handle transients better than analog solutions. This means you need fewer capacitors. That save money and board space so that is a big deal. I remain a little skeptical of non-linear response, after all the PID loop that analog and digital chips perform is already non-linear in a sense—the further away you are from the set point the faster the loop moves to correct it. Yet it is hard to argue with the scope-shots that CHiL has that shows their loop converging faster than an analog loop. I can tell you one headache with these non-linear loops—you can’t use a network analyzer to get a bode plot. They get all helter-skelter since the non-linear stuff kicks in and gives – surprise, a non-linear response. One other downside of digital power is that the chip does have to boot. Larry told me this is about a half a millisecond, but that may matter to people in the non-CPU power fields. The great thing about CHiL is that they have boards and has customers that have already adopted the technology. Even if you do not design PC CPU power systems it is good to pay attention to these new solutions so you can see what might be coming in your field.

One of the things I do find it amusing that where Linear Tech and Analog Devices and many others gave up on CPU power because they felt there were no margins, there is still a high-performance high-margin opportunity in this market if you can make a chip that improves the overall system, even if the chip is doing different things than the old simple analog CPU VID chips. Julian Simon was right, human ingenuity is the ultimate resource.