[SOLVED] Voltage Reference vs. Current Reference

[rmanalo]

Hello everyone,

When designing a temperature-independent voltage reference, the current is PTAT and takes a few extra design steps to get a low-TC current reference. Most applications for a precision voltage reference is on ADCs but how about a current reference? when do we use a CTAT, PTAT, or temperature-independent current reference? all I could think of is biasing mosfets.

best regards,
rmanalo

 

[d123]

Hi,

According to **broken link removed**:"This current reference finds its place in potential applications that include digital-to-analog converters, analog-to-digital converters, phase locked loops, oscillators, analog buffers and operational amplifiers."

I'd have said op amps (and comparators). Some voltage regulators perhaps. A sawtooth (oscillator) in one format/design is just a current source, and that would be hideous if it drifted with temperature.

Current Sources and Voltage References by Linden T. Harrison (get a copy if you can) says on page 87 (pdf page 87) "...Michael Wyatt. The Cascode Peaking Current
Source, depicted here in Figure 4.28, was developed at Honeywell’s Space Systems Division (Clearwater, Florida) by Michael Wyatt during development of a
proprietary monolithic RF IC project in the late 1990s."

I imagine any space and presumably military applications would need a temp-independent current reference.

I strongly recommend taking a look at the truly superb (comparator) current source design for the LM339 in AN-74 LM139 etc A Quad of Independently Functioning Comparators snoa654a if you haven't seen it. Genius (if you can control JFET VGSoff).

I assume some LED drivers incorporate temp-compensated current references/sources, getting off-track a bit.

Don't you think the line/distinction between voltage reference and current reference can be a very blurry one in some senses?

 

[dr_kca]

CMOS thermal sensor circuit is another application in which temperature independent references might be used depending on the used transducer technique.

 

[dick_freebird]

You want to work backward from desired outcomes. The
"care-abouts" for different functions, are different.

Back in the day, some ECL families were designed for a
constant prop delay (tightening up the digital design they
were used for). The internal bias network made it so. The
tempco was set to whatever profile made the delay flat
(probably a somewhat PTAT slope, since in my experience
bipolar always ran slower, hot, even when avoiding saturation).

Another block might want flat gain / gm, a temp sensor would
like the most extreme temp-slope you can make (linearly) and
so on.

If you create a simple (idc) bias master to begin with and then
use parametric analysis to figure out the right behavior at points
away from nominal, that will illuminate what's wanted.