From: Phil Hobbs on
On 12/22/2009 4:24 PM, John Larkin wrote:
> On Tue, 22 Dec 2009 14:55:13 -0500, Phil Hobbs
> <pcdhSpamMeSenseless(a)electrooptical.net> wrote:
>
>> On 8/29/2009 10:42 AM, Jim Thompson wrote:
>>>>>>>>> "Michael Robinson"<nospam(a)billburg.com> wrote in message
>>
>> Since we're on voltage references, I notice that LTC has a really nice
>> CMOS bandgap--the LTC6655--with<800 pV noise (0.01-10 Hz) and 2 ppm
>> tempco in a 5V series reference with 300 mV dropout voltage.
>>
>> That's pretty good going. Now, bipolar bandgaps are horribly noisy
>> because you have to put 20 dB gain on the delta-Vbe before adding it to
>> the diode drop. How come CMOS bandgaps can be so much better?
>>
>> I normally think of CMOS as being much noisier than bipolar at low
>> impedance levels and especially at low frequency, so the LTC6655's
>> performance really surprised me.
>>
>> Any notion of how it works?
>>
>> Cheers
>>
>> Phil Hobbs
>
> ADR421 has been around a while and is almost as good, at around the
> same price. I think it's a jfet thing.

Thanks for the steer. The AD ones run off the delta-Vgs of two JFETs,
running at the same drain current and identical except that one of them
has had two channel implants. Nice idea--buried like a subsurface
zener, so it's away from surface states, and the delta-V is much bigger
than in a bandgap. The TC is smaller too, so less of the (noisier)
compensation voltage is needed.

They don't tell you the details of the temperature compensation, which
is obviously one of the crown jewels.
>
> Vref TCs in the low ppm range require corresponding resistors!
>
Usually at least one, to provide a standard current for an end-to-end
calibration.

> The Intersil floating-gate things are cool.
>
> John
>
Harder to use than they look, though, if you really want the best
performance.

Cheers

Phil Hobbs

(Posting from my new Thunderbird installation with all my old emails,
finally.)