Kenet 12 and 14 bit ADCs
Kenet just sent me a release (pdf) about a remarkable A to D (analog to digital converter) chip. The KAD5512-50 (pdf) ($125/1000) outputs 12 bits at 500 MSPS and uses 370 mW of power. They claim this is almost a decade better than the competition and if they are refereeing to the Analog Devices AD12401 that uses 6.8 Watts (!) to get 400 MSPS well then they are a decade and an octave better. Texas Instruments has the 500 MSPS ADS5463 that uses 2,2 watts. You could take two Linear Technology LTC2241-12 parts that run at 210 MHz and use 585 mW and interleave them to get a 1.17 watt 420 MSPS converter so that is not really a decade worse power than the Kenet part. Still this is a great achievement if the parts are being specified conservatively. Since A to D part specifications are so flexible and demanding this remains to me seen. After all, if the part converts at 500 MBPS but the ENOB (effective number of bits) is only 8, well this is not that big a deal. This Kenet part gives 10.5 ENOB at 250 MHz input frequency and 10.8 ENOB at 10 MHz input frequencies, so don’t forget that analog is analog and something usually has to give in one place to get a great spec in another.
The press release details a family of 12 and 14 bit converters. The 14-bit KAD5514-25 ($89/1000) runs at 250MSPS and uses 320 mW, another best-in-class specification. The also offer lower speed grades for the above converters. The parts run from –40 to 85 degrees. All the parts are pin-compatible, as well as 10-bit pin-compatible parts.
Kenet has been turning up the PR machine the whole year I have been working here at EDN and some claims actually put me off, rather than made me a devotee. The first thing they bragged about was the process. Well, a few questions and it became clear that they are a small fabless semiconductor house and they just buy the same CMOS process that anyone can get. What the technical people were trying to say was that their IP was special and in view of these specs I would have to agree. The next thing Kenet pushed was their connection to MIT and work done there. Well, with so many goofballs in the MIT Media Lab not to mention that professor who claims he can magically couple power into portable devices without induction, MIT provenance is more of a hindrance than a claim to fame. They seem like a bunch of publicity seekers that do some lame things and think we should all bow down. Once again, I was wrong, these specs show that there were some real breakthroughs that probably go back to the MIT DNA. I guess I should not let a few attention grabbers make me think all of MIT is lame. Bob Pease is pretty cool and he went to MIT.
Finally Kenet makes a big deal about how their parts work on charge, indeed, they have copyrighted the term femtocharge©. Well this sounded suspect to me. I think of my pal Henry Ott and how he says that fields make currents, not the other way around. How you conceive of an electronic system working is a very subjective thing. Indeed, voltage is a across variable, current is a through variable and charge is a bunch of electrons sitting there doing not much of anything. So I called up all my IC design buddies and converter experts and the consensus is something like “Look, a data converter is a complex mixed-signal sample-data system. If you want to do the math with voltage that is fine, if you want to do the math with current, that is fine, if for some reason you like to do the math with charge, go right ahead, but remember, it just comes down to transistors.” I would have to agree. The advances made by Intel in power consumption did not come by thinking about the charge shuttling around the processor but rather, by reducing the leakage currents that add up when you use millions of transistors. If you remember sad-sack Transmeta, they took on Intel (smart move huh?) by emulating X86 code and using a RISC machine so they just had fewer transistors and the total leakage was proportionally less. I don’t know whether Kenet uses fewer transistors or clever IP to get such great power consumption but it will be great to get these parts into user’s hands so we can see how they work in a real system. Your thoughts on these issues are welcome, please chime in below.
PS, Kenet hides their datasheets behind an email registration page, the link above should go directly to the pdf if they didn’t get too fancy with snoopy cookies. It reminds me of Volterra, another company that acts like their datasheets are national secrets.