"rickman" wrote in message
...
Flash converters are the only choice for anything above around 1-2 MSPS
last time I looked hard. Maybe they are pushing towards 10 MSPS with
SARS and SD converters. But for anything higher a flash converter is
the only choice.

When did you last look? Two decades ago?

All the ones I've seen (within say 5 years) from about 20 to 500 Msps and
8+ bits are pipelined SAR. Usually with terrible INL for the higher bits
versions, but that reflects their usage: AC circuits, radio (SDR),
ultrasound, etc., where low DNL is priority.

Often, they're also in a series, so you get like, 65-80-110 Msps and
10-12-14 bits. Likely they use the same configurable die for everything
in that series, and burn some fuses during test to implement the highest
spec the chip meets.

They either state what they are (pipelined SAR), or it's apparent from the
timing -- usually LVDS, SDR or DDR output, where a given analog sample
takes N+1 or so clocks to fully propagate through the chip. The INL/DNL
graphs (if provided) also look much like typical SAR graphs; though I'll
admit I haven't seen a wide breadth of ADC types to compare to.

A flash converter could get more data (perhaps more parallel LVDS lanes?),
with less latency (maybe N/M + 1 cycles, where M is the ENOB of the flash
stage), but those don't seem to be common, at least in the ranges I've
looked at.

They often talk about "proprietary compensation stages" or something to
that effect, suggesting that simply churning out a number in the
conventional method isn't as simple when you're doing it at *that* rate.
Maybe this is something they were missing, and hence why flash was the
only option, back in the day?

They're also moving away even from 3.3V logic these days; the 1.8V parts
are something like 1/4 to 1/8 the power for the same specs. And much,
much faster at the top end.

Tim

--
Seven Transistor Labs
Electrical Engineering Consultation
Website: http://seventransistorlabs.com