Was Edison right all along? An investigation of going back to DC power distribution, and what it means for SoC power management

A quite uncharacteristic paper at the Multicore Expo yesterday discussed energy management not as an SoC design issue, but as a societal issue that extends into the SoC domain. The approach, presented by Synopsys Verification Group R/D director Srikanth Jadcherla, was both refreshing in its scope and very insightful in identifying a common thread of systems design that extends from power-generating equipment and high-voltage power distribution networks to SoC architecture.

Jadcherla began with some familiar observations about the massive amount of generating capacity the world is now dedicating simply to the inefficiencies of our electronic stuff. The numbers are shopworn, but so large as to be startling every time I hear them. The normal next step in such presentations is to make a pitch for unplugging equipment when it’s not actually in use, designing more efficient power supplies, and so on. But that’s not the direction Jadcherla was going. He did argue for slashing standby power in electronic devices further, but only on his way to a more interesting point: that we can eliminate all of the no-load dissipation of power supplies, and much of the conversion losses, by simply eliminating AC distribution.

Jadcherla said there were studies suggesting that a wholesale move to DC power distribution in urban areas, where the bulk of the load is electronics and lighting rather than heaters or rotating machines, would actually save significant energy, along with eliminating AC/DC conversion. This argument assumes that a significant portion of new US generating capacity would be in the form of small-scale photovoltaics, which are inherently DC sources, and that much of this capacity would be interspersed throughout neighborhoods and commercial districts, rather than being clustered in rural areas on the land of people who weren’t in a position to object.

Jadcherla’s argument further assumed that rather than a supply-driven grid management scheme, where huge generators are either switched on- or off-line based on overall demand in the grid, the power-distribution network would be demand-based at a fine granularity, with individual electronic systems, appliances and machines predicting their own needs, submitting bids to the network for power, and thus influencing generating capacity directly. This negotiation could be as local as a city block or neighborhood, rippling out to the wider grid if a local neighborhood had surplus capacity or demand.

Nothing really new so far, but now comes the fun part. Jadcherla pointed out that such a scheme could be implemented by surrounding each energy source and sink on the grid with what he called a green wrapper—an interface that matched the internal power bus of the device to the voltage of the network, protected the network and the device from each other, and maintained two-way communications about the devices’ needs and/or capabilities.

The really interesting point was next. Jadcherla observed that such wrappers would be necessary for power stations, the solar-panel array on your roof, your dishwasher, your HDTV and Blu Ray player, and so on. But the concept scales: you could use exactly the same sort of wrappers on individual modules within the HDTV, and individual functional blocks within the SoC within the HDTV. Not only are the functions of connection, protection, and two-way conversation all present and at least similar across the whole range from a nuclear plant to a video CoDec, but it may be possible to create a scalable protocol that would span the bunch. Thus from massive power stations to power domains on a chip, there could in principle be a single hierarchical network for energy management.

Given that there has been precious little discussion of dynamic power-management protocols at the chip level, and given that there is a growing recognition that we need same, Jadcherla’s presentation seems very interesting. At the least it might incite us to look at the experience of the macro-level power grid in implementing demand-based networks—for instance, the horrible experience California had with one such network a few years ago—to see what we can learn.