LM399 based 10 V reference

[fmaimon]

I'm designing a LM399 based 10V reference. It will be used with Richard's (amspire) general purpose power supply, probably the mk3 one.

Anyway, I want to make it with as high accuracy as I can, so I bought these resistors from ebay, in 75K, 32K and 3k11 values: http://www.ebay.com/itm/10-Vishay-TCC-S102C-S102K-Metal-Foil-0-01-Precision-Resistors-variety-value-/170926783729

They are supposedly used Vishay S102K metal foil resistors 0.01% 2 ppm/C. I've measured them with my Agilent 34410A and they are within specs on the resistance, but I didn't measure the tempco.

I also bought some LM399A from ebay and made this:


The PCB layout is this one



What do you think? As I don't have many LM399 and resistors (they are expensive!), I don't want to build one without some feedback.

[free_electron]

throw away that op177... ad706 is the way to go.

much lower input noise.

[ivan747]

You will need to pay attention to a lot of details in the power supply in order to achieve the kind of accuracy you will get out of this reference, if you get it right. High spec op-amps and remote sensing will be needed to remove offsets and line regulation over load, respectively. You will also have to take into account the switching noise (mk. 3 is designed to be used with a tracking pre-regulator). If you don't filter the output right, your accuracy and stability will be down into the noise.

Good luck, I'm looking forward to this project 

[codeboy2k]

Have you breadboarded this circuit?  It looks like a comparator.   How does it even start??

There is no clear starting current to get the zener into breakdown... perhaps the opamp starts hunting from rail to rail at the start and then it gets going? I am not sure. I don't like it this way, since you're not sure if it will start.

You picked an LM399, so you want stability, so let's give it some.  You should have a constant current source on that board to feed the zener some real, reliable current, about 1.5mA or so. The LM399 can (at its worst) change its output voltage about 2mV for every 1 mA change in input current.   The simplest constant current source is a constant voltage source and a resistor. So use your 15V regulated supply and a 5.6k resistor for about 1.45mA, and feed that into the top of the zener, it's not the best design, but it's ok.  I say it's not the best because the 15V can change, according to the LM317 specs.  i.e. it can change because of load, or temperature, or the 2 divider resistors that set the voltage can heat up and change, so your 15V changes too... When it changes, then your zener current will change (slightly), and the reference will change (slightly). How much it changes you will have to do the math If you are concerned about this, then the best love you can give your LM399 is to use another opamp to make a precision current source of 1.5mA and use that to feed the LM399. (there are many examples in almost every datasheet) 

Once you have a constant current, then you can take the reference voltage from the top of the zener and buffer it with a low offset, low drift opamp. You can also use this same opamp to gain up your reference to 10V here, if you wish to.

The LM399 is very stable, but not accurate, so you'll need some way to trim the voltage to your desired reference. That's why I suggested the opamp to buffer and gain it up and at the same time you can trim it to exactly 10V.

The suggestion Free Electron made to use a AD706 is perfect,  low offset, low drift, low bias, no chopper noise, and you get 2 in a single package, so one can be your current source   But to make a current source you'll need another shunt reference, it doesn't need to be the same spec as the main reference, because when it changes there's a much, much smaller effect on the main reference.

[Rufus]

You should have a constant current source on that board to feed the zener some real, reliable current,

His circuit provides a constant current source as stable as the reference.

I would be surprised if it doesn't start the OPA177 can't drive to ground. It probably sources some current while trying to, if not a bit of pull up on the op-amp output will fix that.

And yes if it is supposed to be a 10v reference it needs +/-5% of trim.

[fmaimon]

throw away that op177... ad706 is the way to go.

much lower input noise.

I've choose the OPA177 as I already have lots of it. But why the AD706 is better than the OPA177, in this case? Is it worth to buy the AD706 as the LM399 is have a typical spec of 7uV RMS noise? In both cases they are at least an order of magnitude better.

[fmaimon]

You will need to pay attention to a lot of details in the power supply in order to achieve the kind of accuracy you will get out of this reference, if you get it right. High spec op-amps and remote sensing will be needed to remove offsets and line regulation over load, respectively. You will also have to take into account the switching noise (mk. 3 is designed to be used with a tracking pre-regulator). If you don't filter the output right, your accuracy and stability will be down into the noise.

I'm aware of the noise of the pre-regulator. I'm just concentrating in one piece at a time, so right now it is voltage reference time... 

Good luck, I'm looking forward to this project 

Thank you. Unfortunately, don't hold your breath too much. Lately I'm not getting much free time, so it will take long. 

[fmaimon]

Have you breadboarded this circuit?  It looks like a comparator. How does it even start??

There is no clear starting current to get the zener into breakdown... perhaps the opamp starts hunting from rail to rail at the start and then it gets going? I am not sure. I don't like it this way, since you're not sure if it will start.

You picked an LM399, so you want stability, so let's give it some.  You should have a constant current source on that board to feed the zener some real, reliable current, about 1.5mA or so. The LM399 can (at its worst) change its output voltage about 2mV for every 1 mA change in input current.   The simplest constant current source is a constant voltage source and a resistor. So use your 15V regulated supply and a 5.6k resistor for about 1.45mA, and feed that into the top of the zener, it's not the best design, but it's ok.  I say it's not the best because the 15V can change, according to the LM317 specs.  i.e. it can change because of load, or temperature, or the 2 divider resistors that set the voltage can heat up and change, so your 15V changes too... When it changes, then your zener current will change (slightly), and the reference will change (slightly). How much it changes you will have to do the math If you are concerned about this, then the best love you can give your LM399 is to use another opamp to make a precision current source of 1.5mA and use that to feed the LM399. (there are many examples in almost every datasheet) 

Once you have a constant current, then you can take the reference voltage from the top of the zener and buffer it with a low offset, low drift opamp. You can also use this same opamp to gain up your reference to 10V here, if you wish to.

The LM399 is very stable, but not accurate, so you'll need some way to trim the voltage to your desired reference. That's why I suggested the opamp to buffer and gain it up and at the same time you can trim it to exactly 10V.

The suggestion Free Electron made to use a AD706 is perfect,  low offset, low drift, low bias, no chopper noise, and you get 2 in a single package, so one can be your current source   But to make a current source you'll need another shunt reference, it doesn't need to be the same spec as the main reference, because when it changes there's a much, much smaller effect on the main reference.

I've not breadboarded it, but I'm pretty confident it will work. It is a constant current circuit. Initially, as the OPA177 output can't go to ground (typ. 2V), we will get an initial current of about 0.6 mA to the zener, through R4. It will start it up and get to about 6.9V. This voltage will be amplified by 1.468 to about 10.13V. Then the current will be (10.13 - 6.9) / 3k11 (R4) = 1.04 mA and will be as stable as the LM399.

I'm really not worried on the accuracy of the reference. 10.13 V will be good enough. The output voltage will be adjusted by the circuit described in this post.

Now reading the datasheets, why is the AD706 much better in noise than the OPA177? The typical input voltage noise @ 1kHz BW of the 706 is 15 nV/sqr(Hz) and the 177 is about 8 nV/sqr(Hz), both reading the noise graphs.

[fmaimon]

You should have a constant current source on that board to feed the zener some real, reliable current,

His circuit provides a constant current source as stable as the reference.

I would be surprised if it doesn't start the OPA177 can't drive to ground. It probably sources some current while trying to, if not a bit of pull up on the op-amp output will fix that.

That's my plan. First I will etch a board at home. If it works, I will send it to seeedstudio or itead, else, I'll fix it.

And yes if it is supposed to be a 10v reference it needs +/-5% of trim.

As I said in the other post, I don't really care of the accuracy. I will trim it elsewhere in the power supply. I just want it stable.

[BravoV]

There are 4 purple colored rectangles at the pcb (1 vert & 3 horz), are they cut out holes ?

[fmaimon]

There are 4 purple colored rectangles at the pcb (1 vert & 3 horz), are they cut out holes ?

Yes. I don't want the heat from the regulator getting to the reference circuit.

[ivan747]

But to make a current source you'll need another shunt reference, it doesn't need to be the same spec as the main reference, because when it changes there's a much, much smaller effect on the main reference.

A carefully designed current source based on the LM336 voltage reference should work fine. There's a circuit on the datasheet for minimal temperature coefficient. Tolerance doesn't matter, you will have to calibrate the gain of the buffer op-amp to get 10.0000V anyway. Stability is rated at 20ppm per 1000 hours. With temperature coefficient compensated and this stability, you should have a decent current source for the LM399.

Now we have an LM317 (first reference) feeding a LM336 (second reference) with a constant current, only using a series resistor, and finally the LM339, with a proper current source.

[codeboy2k]

His circuit provides a constant current source as stable as the reference.

I would be surprised if it doesn't start the OPA177 can't drive to ground. It probably sources some current while trying to, if not a bit of pull up on the op-amp output will fix that.

I've not breadboarded it, but I'm pretty confident it will work. It is a constant current circuit. Initially, as the OPA177 output can't go to ground (typ. 2V), we will get an initial current of about 0.6 mA to the zener, through R4. It will start it up and get to about 6.9V. This voltage will be amplified by 1.468 to about 10.13V. Then the current will be (10.13 - 6.9) / 3k11 (R4) = 1.04 mA and will be as stable as the LM399.

Given what you've told me about its startup, then I think I am also convinced it will get enough starting current and will eventually reach regulation.  And I understand that the output buffer drops a voltage across the 3k11 and provides the drive current.  But I see that design decision has problems that are easily remedied.

1) All of the regulating current for the reference comes from the output of the opamp, which is also feeding other circuits in your power supply. You should immediately see this is a problem for maintaining the stability of your reference. This would only be acceptable if you are going to put another voltage follower immediately at J2 before it goes anywhere else on to the rest of your circuits. And if so, why not put that buffer on the reference board itself, so that J2 is the output of a real buffer, not the output of your self-referential voltage reference that also provides it's own current source.

2) R4 provides a path for any voltage or current noise from outside your reference board to get in and throw the reference off course.

3) if J2 is feeding outside, and something outside sinks a large current, say an ADC sample and hold event, then it will steal current that is supposed to be for your constant current source to your reference. I don't know how you can call it a constant current source when you are taking that node off board and allowing other circuit nodes to siphon current away from it. That net feeding your LM399 should be stable and quiescent. Yes, I realize the opamp will provide more drive current when needed, but it cannot recover instantaneously, and do you really want the junction at the top of your current source changing that much due to dynamic loads off-board? I would not.

4) I don't like that your reference has no actually set current.. you really don't know what it will be, just trust that it will stop rising in voltage value as the opamp gives it more and more current until it settles.. and yes, I understand at that point it will be stable. But in my opinion I'd like to actually SET the current to something I specify, and maintain it there.

My design would have a separate op amp for the current source and one more as a buffer. As I said, if your going to the effort of using an LM399, why do something that can mess up your reference?

Actually, I'd guild the lily, and use the second opamp not for merely a current source for the main reference, but as a servo for the LM317 providing your 15V rail.  servo the LM317 to stabilize it over time and temperature, and then you can set your 1.5mA  current reference with a 5k42 low tempco resistor through the LM399. Use the original opamp as a  buffer amplifier where you can dial it up to 10.0000V

Now reading the datasheets, why is the AD706 much better in noise than the OPA177? The typical input voltage noise @ 1kHz BW of the 706 is 15 nV/sqr(Hz) and the 177 is about 8 nV/sqr(Hz), both reading the noise graphs.

Actually, I'm quite familiar with the AD706 so I know it's a good amp and there's something to be said for going with what you know.  But I just read the datasheets for the OPA177 too, and I agree, it seems quite a lot better in noise, offset and drift specs, and is designed to replace the AD707, which is an even better amp then the AD706.

The only thing the AD706 has over the OPA177 would seem to be its sub-nanoAmp bias currents, which you don't need to worry about. So your choice of OPA177 will work good here.

Sorry for the long post.  I get windy.

[kt315]

I am considering building Richard's PSU as well. I do not have a lot of experience with electronics, so excuse my ignorance ...

What are the advantages/disadvantages of using parts like LTC6655B or LT6654A comparing to LM399?

Also, which parts precision would affect accuracy of the reference circuit the most? My current understanding, is that it is critical to get the apm gain right on the reference, hence the divider accuracy would be critical.

Would not it be a problem for precision if the reference is in fact a separate board?

EDIT: In a mean time, off to ebay I go to buy some fancy TFT display ...

[free_electron]

Zeners are current driven objects. The input cuurent noise is in the femto ampere range for an ad706 and i. The picoamp range fro the op177. Furthermore the ad706 is dc stable. This opamp is built for dc operation. The input bias current is also stable over temperature, that of an op177 is not.

This drift in n put bias current willl shift the operating point of the zener ever so slightly.

Remeber that you are fishing for digits behind the comma here. So the question is: do you want a 10.0000 volt reference , i. Which case the op177 will work. Or do you want a 10.000000 reference that is temperature stable . In which case you will need the ad706...

[codeboy2k]

One of the great things about Dave's forum here is that there are great discussions that get one thinking..

I was curious about this discussion, the AD706 vs the OPA177.. So I did some math. I've attached images and Smath Studio output for you viewing pleasure. 

Summary: Even though the initial offset of the AD706 is 100uV and the initial Vos of the OPA177 would seem better at 60uV, the actual output error is dominated by the bias and offset currents through the parallel combination of the (Rfeedback||Rinput), to the point that the total initial error due to offset voltage, offset current and bias current is smaller with the AD706.  Furthermore, a 20C rise in ambient temps will change the output of the OPA177 by ±11.6mV (±6.7mV RMS) ±179uV (±106uV RMS) but will only change ±45uV (±44 uV RMS) when using the AD706.  I did the same with a 5C rise, and the OPA177 will change by ±2.9mV (±1.7mV RMS) ±45uV (±27uV RMS) and the AD706 will change by only ±11uV (±11uV RMS)

Conclusion.. The AD706 is clearly more temperature stable, and its lower input currents even help in reducing total output offset errors.  However, the difference is about 130uV over a 20C change, and a mere 30uV at 5C.  Reducing the feedback and input resistors while maintaining the desired ratio can improve this.  Also, lower input resistors lowers the input noise as well.

And as Free Electron said, the higher bias and offset currents of the OPA177 will ever-so-slightly change the zener reference setpoint, and so that additional zener change will be added to the offset error, and will make the output offset error of the OPA177 even higher than I calculated here.  The AD706 will affect the zener much less.

Also, the OPA177 has better noise specs than the AD706, but I did not do noise calculations. Given the closeness in temperature performance, especially at low temperature changes (ie. when the environment of the reference might be at a stable temperature), and that this can further be lessened by smaller valued resisters, then  perhaps the OPA177 might be a better choice for its noise spec alone, if the error budget allows for the 10s of uV changes due to temperature.  You would be able to achieve 10.0000XX. The AD706 would give one more digit, at 10.00000X

The images attached are my results.  The PDF is the same as the images, but in one PDF file. The ZIP is the source of the Smath Studio input file. Smath Studio is a pretty decent free Mathcad clone and I used it to do all the calculations here and print the images.


[edit: updated the attachments to correct the mistake re: OPA177 Ib and Ios]

[fmaimon]

Given what you've told me about its startup, then I think I am also convinced it will get enough starting current and will eventually reach regulation.  And I understand that the output buffer drops a voltage across the 3k11 and provides the drive current.  But I see that design decision has problems that are easily remedied.

1) All of the regulating current for the reference comes from the output of the opamp, which is also feeding other circuits in your power supply. You should immediately see this is a problem for maintaining the stability of your reference. This would only be acceptable if you are going to put another voltage follower immediately at J2 before it goes anywhere else on to the rest of your circuits. And if so, why not put that buffer on the reference board itself, so that J2 is the output of a real buffer, not the output of your self-referential voltage reference that also provides it's own current source.

2) R4 provides a path for any voltage or current noise from outside your reference board to get in and throw the reference off course.

3) if J2 is feeding outside, and something outside sinks a large current, say an ADC sample and hold event, then it will steal current that is supposed to be for your constant current source to your reference. I don't know how you can call it a constant current source when you are taking that node off board and allowing other circuit nodes to siphon current away from it. That net feeding your LM399 should be stable and quiescent. Yes, I realize the opamp will provide more drive current when needed, but it cannot recover instantaneously, and do you really want the junction at the top of your current source changing that much due to dynamic loads off-board? I would not.

I really don't see much problem, as long as the output current is within the opamp limits. If I recall correctly, not even the 3458a has an output buffer in it's reference board. But, as I will put an buffer after the board, there is no harm in relocating it.

4) I don't like that your reference has no actually set current.. you really don't know what it will be, just trust that it will stop rising in voltage value as the opamp gives it more and more current until it settles.. and yes, I understand at that point it will be stable. But in my opinion I'd like to actually SET the current to something I specify, and maintain it there.

I don't follow you there. The current is set by R4. Why are you worried about the turn-on transient current if it every circuit will have the same problem? Even the best constant current circuit will do exactly that as it turns on. In the end, it won't matter, as I intend to keep it on as long as I can.

Actually, I'd guild the lily, and use the second opamp not for merely a current source for the main reference, but as a servo for the LM317 providing your 15V rail.  servo the LM317 to stabilize it over time and temperature, and then you can set your 1.5mA  current reference with a 5k42 low tempco resistor through the LM399. Use the original opamp as a  buffer amplifier where you can dial it up to 10.0000V

The circuit, as you drawn is pretty much the same as my circuit, using the LM317 as a high output current buffer. If you put another buffer in there, you will loose that.

One of the great things about Dave's forum here is that there are great discussions that get one thinking..

I agree with that! 

I was curious about this discussion, the AD706 vs the OPA177.. So I did some math. I've attached images and Smath Studio output for you viewing pleasure. 

Where did you get those bias drift numbers for the OPA177? The datasheet, page 2, lists 85 pA/C Ios drift and 60 pA/C Ib drift, a lot worse than AD706 but much better that the ones you've used.

[codeboy2k]

I don't follow you there. The current is set by R4. Why are you worried about the turn-on transient current if it every circuit will have the same problem? Even the best constant current circuit will do exactly that as it turns on. In the end, it won't matter, as I intend to keep it on as long as I can.

Yes, I realize R4 sets the current, and it will be as stable as the LM399 itself is.  I'm not worried about the turn-on current per-se, but that you don't know what your current through R4 will be, exactly, in the end. The feedback from the opamp output as it rises to 10.XXX volts will provide a rising current through R4 to the LM399. Once the rising current stops making big enough changes in the zener's output, then the opamp output will also stabilize at 10.XXXX something volts.  I clearly understand this. I think because your application doesn't care about the final voltage, that you can accept it this way. You said you would correct it in the PSU software.

The circuit, as you drawn is pretty much the same as my circuit, using the LM317 as a high output current buffer. If you put another buffer in there, you will loose that.

I only provided a way to stabilize the +15V power rail from the LM317, using a 2.5V reference instead of the 220 ohm voltage dividers to the ADJ pin.  A stable reference and opamp to servo the ADJ pin.  Thus the +15V power is now stable over time and temperature, and can then be used to provide stable current to the LM399.  At the same time, the rest of your reference circuit would benefit from a more stable power rail.  Its probably not necessary in your application, as I realize you aren't concerned that the final voltage isn't exactly 10.0000 and so you actually have a much more stable current than anything else, since you're using the LM399 itself to provide it's own current source. Just don't let anyone (anything!) take away current from that output, so it's good that you will buffer it.

Where did you get those bias drift numbers for the OPA177? The datasheet, page 2, lists 85 pA/C Ios drift and 60 pA/C Ib drift, a lot worse than AD706 but much better that the ones you've used.

Oops.. I'm cross-eyed. mistake. my bad. I picked up the numbers I used from the line above the real numbers that you quoted. Yes, the real numbers are 85 pA and 60 pA/C.  Thanks for catching that... I'll fix it and re-upload the pages.

[branadic]

Hi,

a few weeks ago I've build the circuit presented in the datasheet of the LM199/399 (Linear Technology) called "10V Buffered Reference" twice.
The references where pre-measured to find the best fitting resistor values. The resistors right now are smd types with 5ppm/K, contrary to the pictures attached.
For stability of them I thermo-glued a LTCC-board-heat-regulator on top of them (http://www.kuhne-electronic.de/en/products/crystal-heater/qh-40-a.html). This solution is much cheaper compared to 1ppm/K resistors and was faster to solve.
BTW: Resistors can be bought over here http://www.rhopointcomponents.com/components/resistors/precision-through-hole.html
The LM399 is packaged into a thick walled plastic package on both pcb sides without contact to the slot arms to keep air flow away.
Both boards have know seen a 300h burn-in @ 120°C and output 10,0019V and 10,0034V powered from a lab power supply (stable for more than a week, poorly measured only with a 34410A because of missing a 3458A or null voltmeter) without change in voltage under lab conditions.
You said you want it stable, then take humidity into account too. I've used the LT1001 in CerDIP package sitting in a precision socket.

[ddavidebor]

wow, amazing design!

[codeboy2k]

Those are nice boards you built.  The heater you linked to is pretty good, it claims to regulate the temperature to with 0.1C

If a reference is made with that heater, and also heating the opamp and resistors, and could keep it all within 0.1C, then it will be really stable, as you've seen.  With a 0.1C stable temperature, the OPA177 from this thread will only change between  500-900 pV at the output.  The AD706 is still better, will change only 200pV.  But both opamps should be stable to 1uV, and give 10.000000, and for sure able to hit 10.00000X (i.e. 10uV, and the 1uV digit would be uncertain)

[branadic]

The boards are nothing special, just my expermiental environment with the LM399.
BTW: Current Sources & Voltage References by Linden T. Harrison is very nice "evening-reading".
You can read that the LM199 was designed by Robert Dobkin and introduced in the early 1970s by National Semiconductor, the predecessor of the LM299/399/3999 which had less stringent specifications and covered shorter temperature ranges. He also writes about 90°C chip temperature, that it requires 7,5V to operate the reference and more than 9V for the heater.

Nevertheless, the 10V are impractical in most todays applicatons and a voltage divider (done in the 34410A to produce also 5V out of the 6.95V of the reference) will introduce further tempco and noise. You can overcome the tempco by using a resistor network or expensive low / zero tc resistors, but this is also a bad way.
I agree that the heater is very interesting and as mentioned the LTC6655 but also the LT1236 have perfect characteristics when using them in the LS8 package. Combined with a heater/oven that can give a more practical to use reference voltage for adc or dac applications.
So this is the idea I'll follow next. I've two LTC6655BHLS8-2.5, a few LT1236AILS8-5 and two further heaters.
Voltage references, to be continiued... 

[codeboy2k]

yes, I agree 10V as  a reference inside an instrument is not really necessary anymore.  There are so many good references that it's easy to find one off the shelf that fits and is accurate enough for the task at hand, depending on the error budget allowed.

This volt-nut stuff is (in my opinion) for something on the bench, i.e. a 10V stable benchtop reference that I would use with a K-V divider for research  and prototyping. Or as a stable reference for a precision current source down to the 100 uV or so.  I realize that the OP is using this reference in a precision PSU. 

I don't think I can use a PSU that needed a reference this stable.  If I am building a PSU I would want it settable to 1mV resolution at the best, and probably 10mV is good enough, for my needs. Obviously others have different needs.

My current favorite plug and play reference that can be installed and doesn't need calibration or trimming is the MAX63XX series, with the best chip having 0.01% initial accuracy and 1ppm/C drift. It's expensive, like the LT6655, but I think it's a little better too.

Have you ever used Intersil references? I have not, so I was curious about their FGA technology compared to the buried zeners and bandgaps.

[free_electron]

throw away the socket... you can get electrical noise due to stress in the socket....

[branadic]

I was looking forward for some MAX6350MJA, first one's sold on ebay, but tests showed that this were plagiarism. Second time I bought them via a distributor and again got plagiarism, but this time I was sure I could send them back and get back my money.
I don't want to use the plastic package also because of humidity influence and related stress.
Maxim couldn't help me with parts as they don't stock them by now and I would have to buy a dozen oft them (I guess they wrote 100.000 or more in the mail) just for a few one that I need.

throw away the socket... you can get electrical noise due to stress in the socket....

This is a typical "once heard" statement used as an answer for all questions. Stress is less a problem also because of the CerDIP package, air flow is much more critical. But you seem to forget the LT1001 is not the reference but the buffer amp, so I don't care about your "electrical noise due to stress in the socket" argument.

However, there are a few other voltage reference ics in my lab like AD584TH, MAX6126AASA25+, MAX6126AASA50+, AD586LQ, LM334H, REF02DT, ADR444ARZ/ARMZ, enough material for further projects. All that is missing is a good old LTZ1000ACH that I'd like to have as an already used/burned-in board.

BTW: There are a few other threads concering references in eevblog, e.g.:

https://www.eevblog.com/forum/chat/5v-precision-referene/
https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/
https://www.eevblog.com/forum/projects/voltage-reference/

And again, there are several links to ready to use boards:

http://www.gellerlabs.com/Voltage%20References.htm
http://www.voltagestandard.com/