6 ms·
MIT Startup Aims to Make Data Centers Efficient with Power Conversion Tech
- kwantam 15y agoThis article is really light on detail. A quick search of DSpace, MIT's paper/thesis repository, brings up the following paper from Giuliano in 2009: "Fully monolithic cellular buck converter design for 3-D power delivery" http://dspace.mit.edu/handle/1721.1/61350 http://dspace.mit.edu/handle/1721.1/61350 It's not clear that it's what they're trying to productize, but it's likely at least related.
- aidenn0 15y agoI'd like some more details; they must be doing something very novel if they are eliminating the inductor. If you just build a traditional switching converter with a capacitor instead of an inductor, you end up with less than 50% efficiency.
- Daniel_Newby 15y agoTheir apparent patent: http://www.google.com/patents/about?id=_wPKAAAAEBAJ http://www.google.com/patents/about?id=_wPKAAAAEBAJ It appears to be a switched capacitor front end for a conventional inductor-based converter. The special sauce appears to be that the design requirements for the back end converter are relaxed if its voltage varies over a smaller range. Efficiency can be much higher than 50% if the capacitors are only lightly discharged during a cycle. However under high load that means fast switching, which means high capacitive losses in the switch transistors. A hybrid approach might be a big win for small systems (cell phones) that spend most of their time asleep, and it could save cost and circuit board area over a complicated SEPIC converter. Their elevator pitch: http://www.thebusinessmakers.com/index.php?id=2761 http://www.thebusinessmakers.com/index.php?id=2761 Ouch. Data centers are not small systems that spend most of their time asleep. My gut feel is that even the static current draw from a big CPU and its DRAMs is well outside the sweet spot of this technology.