Low drift OP amp for photodiode circuit

Opamps with very low input currents have been available for a very long time; you don't really need the latest part for that task, unless you want to operate at low supply voltages, with very low supply currents, etc.

One class of opamps well worth mentioning are NSC's inexpensive CMOS opamps, e.g., their LMC6001A. This one is 100% tested to insure that each chip has input leakage currents of less than 0.025 pA, or 25fA, as we like to say. These have a reasonably-low input offset voltage of 350uV max.

Note, the LMC6001A was introduced at least 12 years ago (I have a 1995 datasheet in my computer's collection).

Normally if you want to measure very low light levels (very small photodiode currents) you'll just use a high- value feedback resistor to develop a signal voltage above the opamp's offset voltage. For example, with a modest 100G resistor you can measure down to say 5fA and develop a 0.5mV output signal. Using the LMC6001A, which has up to 0.25mV of offset voltage and up to 2.5mV developed from input leakage current across 100G, you could improve your measurement range by noting the dark value and observing the difference when you turned on your weak light.

Note, 5fA of photodiode current would mean you're observing only 10fW of light, using a 0.5A/W silicon detector rating.

If you can chop your light, you can go even lower with a lock-in amplifier.

If you're not inclined to use a high-value transresistance feedback resistor like 50 to 200G, but instead want to use say no more than 1000M, you could use a CMOS chopper-input opamp with 5uV or less of input offset voltage. There have been some interesting new parts in that area.

Since ultralow light level measurements are so slow anyway, one nice method is to use a capacitor instead of the feedback resistor, and integrate the charge coming from the PD. You can reset the cap with a MOSFET wired in parallel with it (there are other methods too--my favourite uses LEDs).

There are lots of nice things about this method. One is that it's much much quieter at very high gains--unlike 500G ohm resistors, capacitors don't have thermal noise. Thus the slope of the V vs T curve can be measured much more accurately than the instantaneous voltage across the feedback resistor. (*)

Another one is that you can do range switching by just changing the frequency of the reset pulses. A third is that you can add a Schmitt trigger and make an oscillator--a current-to-frequency converter--that will make the measurement much easier, and if you have a universal counter, there's no problem getting lots of digits on a frequency measurement.

Of course the best thing to do is to use brighter light. ;)

Cheers,

Phil Hobbs

(*) The reset process does have thermal noise, but as long as you measure the voltage just after reset and subtract that from the measurement, there's no thermal noise in the photocurrent measurement.

Chopper stablized op-amps can be used to bring the drift down nearly to zero. Usually, it is the bias current and not the offset voltage that matters in a photodiode amplifier.

You may want to look at the LTC6078 if you are measuring a near DC light signal. Both its offset voltage and bias current are good. Its bandwidth and Vcc range sucks however.

One trick is to just briefly collapse the opamp power supplies. The esd diodes are there anyhow, so you may as well use them.

One reason I like picosecond electronics, as opposed to temperature or low-level measurement, is that it doesn't involve a lot of waiting.

John

Analog Devices has a nice application book explaining the photosensors in the details. In the short, you need to build a transimpedance amplifier using a low noise CMOS opamp with femtoamp input currents. There is a big tradeoff of sensitivity vs speed.

With the modern APD sensors, you can count virtually every photon of light.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

Even several decades ago, uA725 and OP-07 amplifiers were as accurate on DC drift as could be easily measured. Thermocouple effects from wiring are likely to be dominant in your application, not amplifier drift.

Input current errors are more likely than offset voltage errors to be the limit on a low-light photodiode, so a FET input may be best. Maybe LMC6061? Be careful to shield the input wiring.

How's this? Since there's nothing to be learned during the integration period, disconnect the amp.

v+ | | pd shielded- | coil | reed | relay +-------o o------+------amp---adc | | cap | | discharge | fet gnd | gnd

That's basically a poor man's current dispenser, which is another decent method. I'd be a bit worried about the timing uncertainty in the relay dominating the noise, but the OP probably isn't doing measurements at that level anyway.

Cheers,

Phil Hobbs

wouldn't a photo multiplier tube fit the bill? that is what we use at work to detect extremely low levels of light in the vessels so that the system can shut down at the first hint of problems due to any form of arc.

--
"I\'m never wrong, once i thought i was, but was mistaken"
Real Programmers Do things like this.
http://webpages.charter.net/jamie_5

A 500G resistor has about 0.18fA/rtHz noise, that's about as much noise as a

0.1pA (leakage) current has intrinsically. So if your opamp is (and needs to be) better than that, use the integrator method.

How does the LED reset method work? Using the LED as a photoswitch?

I also like the supply reversal method conceptually but have never used it.

robert

Use a DPDT portion of a 4053 cmos switch with say 2.5k resistors, to reverse the voltage across one of NSC's femptoamp CMOS opamps. The opamp's input-protection diodes and output MOSFET diodes form a balanced bridge, nicely shorting the integration capacitor.

2.5k 4053 ,---/\\/\\---o-->o------+--- +5 1pF | | o---, | ,---||---- | ----, | | | 5k _|_ | | | | ---+--/\\/\\--|+ \\ | | | | >--+--- out | | gnd ---|-__/ | | | 2.5k | | | LMC6061A '---/\\/\\---o-->o---+--| --- -5 o------'

The 4053 has built-in level shifters so even tho it's powered from +/-5V it takes a 0/+5V logic command. The LMC6061A opamp costs $2.11 qty 1 from DigiKey.

It's a bit more complicated but very low leakage--comparable or superior to an unprotected MOSFET, i.e. below 100 fA from -5V to +0.5V bias.

It's just a two-LED diode clipper--when reset is asserted, A2 forces the output of A1 to equal its input, within an offset of I_in R1 (inconsequential in my case). You have to choose R1 and C1 so that the composite amp is stable when RESET is active, but there are other ways of doing that, e.g. running A2 at a higher noise gain. D1/R3/R4 forces the LEDs to be biased off reliably during integration. There's a bit of a hold step, but we need to do correlated double sampling to get the noise improvement anyway, so that isn't a big item.

(In my circuit, I actually use a BJT inverter instead of the extra reset line + diode, and the charge was being dumped into the integrator by another LED switch rather than being continuously integrated. This is from the second generation of my Footprints sensor, which was a $10 thermal infrared imager with 100 pixels and ~0.1K NETD. The second version might have been $20, plus another $40 for the PIC and the box--the system is about 15 dB lower in price than the next lowest, which doesn't work as well.)

Cheers,

Phil Hobbs

D1 |\\ | R3 RESET o-----| >|-------RRRRR-----------------------------+ |/ | 1/4 LMC6034 | (+5 = reset, | 0 = integrate) /| | LED 1 LED 2 / | | / +|-------+ R2 | /| | /| / | | RESET' o---+----RRRRR--|< |--*---|< |------< | A2 | | \\| | | \\| \\ | R (+5 = integrate, | \\ | R 0 = reset) | \\-|--+ R R4 | \\| | R | | R *---------------------+ | R | R1 R || C1 | +--------||------+ | | || | | | | | I_in | |\\ 1/4 LMC | | | | \\ 6034 | | o------*--|- \\ A1 | | | \\ | | | >--------*---------+--O | / Output | / +----´+/ | |/ | ----- \\ / V

Should be R1 R || C1 | +--------||------+ | | || | | | | | I_in | |\\ 1/4 LMC | | | | \\ 6034 | | o------*--|- \\ A1 | | | \\ | | | >--------*---------+--O | / Output Vbias | / (2.7V) O----| / |/

(Forgot it was running +5/0 supplies)

Cheers,

Phil Hobbs

It's worse than that though--don't forget the big noise peak caused by the input capacitance and the amplifier's input noise--with a 500G resistor and a 3 pF input capacitance, the noise starts to rise linearly at 1 Hz! That doesn't exist with charge sensitive amps. And also the bandwidth can be quite different. If you use a capacitor across Rf to reduce the bandwidth to the same value you'd get by charge dispensing, you basically wind up with an integrator with a continuous resistive reset as opposed to a switched one.

Cheers,

Phil Hobbs

Yeah, that's a problem. Then there's the e_n/I_leak tradeoff with JFET inputs. I haven't looked into CMOS amps in the past 10 years; have they improved significantly noise-wise?

I wonder if the switched-integrator method could be beneficial in an all-analog signal chain; i.e., if the sawtooth output could be differentiated back into a voltage so that the overall noise/bandwidth performance would be superior to what could be practically achieved with a transimpedance amp.

robert

Yes of course. In fact that's a standard approach for patch-clamp amplifiers. But you do need a bandlimiting resistor for the differentiator, like this addition to the integrator-reset circuit I posted:

2.5k 4053 ,---/\\/\\---o-->o------+---- +5 C1 1pF | | o---+--| --- -5 ,---||---- | ----, | | R1 | 5k _|_ | | | ,---/\\/\\---, ---+--/\\/\\--|+ \\ | R2 C2 | | | ___ | | >--+---/\\/\\--||-| -| --+--|- \\ | gnd ---|-__/ | | | >--+--- | 2.5k | | | ,--|___/ LMC6061A '---/\\/\\---o-->o---' | | o------' gnd

The transimpedance gain is Rf = R1 C2/C1, e.g. Rf = 1000G for R1 = 10M and C2 = 1000pF. R2 sets the differentiator bandwidth, e.g. 50kHz for 3.3k in this case.

The third 4053 switch can be used to disconnect C2 from the output stage and discharge it during a C1 capacitor reset. Properly wired, this switch can include an output hold function.

I had my technician build me one of these, does anybody wanna know how it works?

sure, shoot

Le Sat, 29 Sep 2007 07:57:13 -0700, Winfield a écrit:

LOL! Sure we do :-)

--
Thanks,
Fred.