EEVblog #1318 - What's State-of-the-Art in µCurrent Opamps?

[EEVblog]

Is there a better chopper/auto-zero amplifier than the MAX4239 used in the uCurrent?
A look at the exciting world of parametric searching for components. Follow along as Dave looks for a new state-of-the-art opamp. Success not guaranteed!

[eV1Te]

I agree with you that it's a lot of fun to research new parts like this.

I think current-sens amplifiers might be the wrong category, because you are not measure currents and you are not interested in low input bias currents etc.
In your application you are measuring low voltages, hence try looking into (and please make a video) on nano-voltmeters! Keithley have some nice documents on this, and you should be able to find circuit diagrams of some of the older models.


On a related topic, I recently developed a fC charge amplifier for DC to a few Hz. The best I could find for such applications was the impressive LMP7721 with 3 fA bias current and 26 uV offset voltage (very low for such an amplifier). It is easy to forget that burden/offset voltage is very important for low current measurements as any offset allows for voltage differences which causes leakages, even with guarding.


Edit:
Of course nano-voltmeter type circuits are usually only for low frequency so perhaps not applicable, but interesting nevertheless.

Another option would be transfer impedance amplifier circuit (at least for 10 mA and below) = Close to zero burden voltage at any current and could easily work down to pA currents (with reduced bandwidth).
I found this article from TI where they design a 1 mV/nA amplifier with 100 kHz bandwidth powered from +-2.5 V rails (very close to the lowest range on your uCurrent):
https://e2e.ti.com/cfs-file/__key/telligent-evolution-components-attachments/01-930-00-00-00-66-60-61/Op-Amp-Bandwidth-for-Transimpedance-Amplifiers.pdf

[ali_asadzadeh]

Check this low cost part,TP5552 from 3PEAK, it's around 5uV but it's under 0.6USD for 10 units.
https://lcsc.com/product-detail/Others_3PEAK-TP5552-VR_C248604.html

[Whales]

Check this low cost part,TP5552 from 3PEAK, it's around 5uV but it's under 0.6USD for 10 units.
https://lcsc.com/product-detail/Others_3PEAK-TP5552-VR_C248604.html

The offset voltage distribution graph has a weird horizontal scale, but otherwise suggests the part is much better than the max of 5uV. 

I saw some other interesting parts by 3peak on LCSC once.  Their datasheets seem easy to read (I think they're copying some of the better layout styles), but I don't know anything more about them.

EDIT: Main website: http://www.3peakic.com.cn/En
EDIT2: Their low drift parts seem to have low Voffsets, but their  low noise parts don't.  Categories are probably arbitrary anyway.

[Dr. Frank]

The AEC-Q 100, 101, 200 standards describe, how electronic components (IC, discrete and passive components) have to be qualified /tested by the manufacturers before they are allowed to sell their parts to the Automotive Electronics industry, like Continental, Bosch, Valeo, Visteon, and so on. The tests include environmental, reliability and electrical tests.
These are the de facto automotive industry standards, i.e. very important for us. Go to www.aecouncil.com for details, and check the 3 base documents.

Specially for Automotive, manufacturers specify and test for -40.. +85°C for Interior, and -40 .. +125°C for Power Train, Body Control and Safety applications, which you already found in all of the chopper specifications.

Use of the standard Industrial range of 0..+70°C is mostly not allowed in Automotive.

This whole subject about electronic components in Automotive electronics, tests, reliability, Quality Systems, changes and obsolescence, and so on might give enough content for an interesting EEVblog.

Frank

[Kleinstein]

If bandwidth is important, one could change the circuit from the 2 x10 stages to one compound amplifier with a 2nd non AZ inside the loop of the AZ OP. So the other OP could be some 20-50 MHz GBW and do most of the gain.
Another advantage would be saving 2 precision resistors for the gain.

[splin]

I appreciate a large part of the video is to show ways to go about finding suitable parts and understanding datasheets and the actual devices chosen and specs are less important.  However I think your concentration on offset voltage is a case of not seeing the woods for all the trees.  The MAX4239 seems to be a rather poor choice  for the ucurrent - it's far too noisy at 30nV/sqrt(Hz). Given the ucurrent's specified 300khz bandwidth, input noise is 108uVpp!  (This is inline with the 10mV output noise reported in these Forums).  Offsets below, say 10uV, are lost in the noise.

The ADA4522, ADA4528 and OPA189 are better at around 5 to 10nV/rt(Hz) but the noise varies with frequency - the 4528 having a large noise peak at 200kHz. The new ADA4523 is better still with 0-300kHz integrated noise of 20uVpp and an outstanding 88nVpp 0.1 to 10Hz. You missed this one because the maximum offset is 4uV, but if you look at the distribution graph, all 160 samples were within +/- 0.5uV offset at a supply voltage of +/- 2.5V. The 4.5mA supply current is perhaps too high for a coin cell supply (but the second amp can be a lower power device as it's noise and offset contributions are only 1/10 of the first). The ADA4523 is cheap at $0.85 @1k.

None of these alternatives were available at the time the current was designed, so the MAX4239 may well have been the best choice for the circuit architecture chosen,  but a better option would probably have been to use a hybrid amp using a low noise fet amp with a zero drift/chopper for DC stabilization.

For DC applications, such as using the current with a slow DMM, most of the noise is filtered out but again the MAX4239 has a LF noise spec of 1.5uV. Added to the 2.5uV max offset spec means that lower noise, but higher offset parts may be as good if not better.

Another problem I have with the video was the discussion of the distribution charts. These are typical characteristics showing parts from early production.  I don't know what real value, if any,  these provide. They aren't guaranteed and any tweeks or changes to the production process can render them useless. You would be foolish to rely on them except perhaps for choosing a part that you will routinely screen as part of your manufacturing process.  Even then the supplier could change the process such that the yields of parts close to the typical values you were relying on become so low as to render your product unviable.

Data sheets often have serious inconsistencies between typical values listed in the specifications and those shown in the typical graphs and some manufacturers have a reputation for supplying parts that almost never come close to their typical values.

Finally, the GBW isn't the whole story -  the ADA4522 has a GBW of 2.7MHz but a closed loop bandwidth at a gain of 1 of 6.5MHz. But why does the ucurrent only have 300kHz bandwidth from the two x10 6.5MHz GBW amps?

[David Hess]

I appreciate a large part of the video is to show ways to go about finding suitable parts and understanding datasheets and the actual devices chosen and specs are less important.

Like Dave apparently, I really enjoy doing parts selection and comparison and how it affects design.  I often end up making big spreadsheets comparing parts with datasheet specifications, calculated specifications, and measured specifications.

However I think your concentration on offset voltage is a case of not seeing the woods for all the trees.  The MAX4239 seems to be a rather poor choice  for the ucurrent - it's far too noisy at 30nV/sqrt(Hz). Given the ucurrent's specified 300khz bandwidth, input noise is 108uVpp!  (This is inline with the 10mV output noise reported in these Forums).  Offsets below, say 10uV, are lost in the noise.

I was going to say the same  thing.  There is a place for a chopper stabilized amplifier in this application, but the design has some contradictions.  See below.

None of these alternatives were available at the time the current was designed, so the MAX4239 may well have been the best choice for the circuit architecture chosen,  but a better option would probably have been to use a hybrid amp using a low noise fet amp with a zero drift/chopper for DC stabilization.

That is exactly what should be done unless the bandwidth is significantly limited, which contradicts using two operational amplifiers in series to boost bandwidth. That is the contradiction I mentioned above; the high broadband noise from the chopper stabilized amplifier limits the usefulness of higher bandwidth.  Using a chopper stabilized amplifier alone is less expensive though.

If it were not for the 10 kilohm source resistance at the highest sensitivity, then a low noise bipolar part would be suitable but in this case, a low noise JFET part, or discrete JFET or MOSFET input stage will allow for almost 2 orders of magnitude better sensitivity while maintaining the same drift by using the chopper stabilized operational amplifier to remove drift and flicker noise.

For DC applications, such as using the current with a slow DMM, most of the noise is filtered out but again the MAX4239 has a LF noise spec of 1.5uV. Added to the 2.5uV max offset spec means that lower noise, but higher offset parts may be as good if not better.

I laughed at that "low noise" claim on the datasheet.  Chopper stabilized amplifiers are only lower noise than the alternatives at much lower frequencies.  The datasheet for a chopper stabilized amplifier lists noise from 1 to 10 Hz only for comparison purposes for comparison with other chopper stabilized amplifiers.  What matters in practical applications is noise from DC to 1 Hz so if this is not listed, beware.

Data sheets often have serious inconsistencies between typical values listed in the specifications and those shown in the typical graphs and some manufacturers have a reputation for supplying parts that almost never come close to their typical values.

Datasheets sometimes outright lie, and often mislead as with the "low noise" part mentioned above, or listing offset in microvolts rather than nanovolts as Dave noticed.  Texas Instruments has a history going back decades of doing that sort of thing but they are not the only one.

[Kleinstein]

The high bandwidth of the max4239 is of limited use because of the high noise - though there are more noisy ones. At higher frequencies the max4239 is not that bad for an AZ OP with supposedly some 30 nV/sqrt(Hz). Still a scope input amplifier may be lower noise for the higher frequencies, or at least comparable.

There is a difference between classical auto-zero OPs (e.g. the max4239, LTC1052, TLC2652) and real chopper stabilized OPs (e.g. OPA189). They have a different noise spectrum. For a high bandwidth design the copper signals in a copper stabilized OP can be a real problem.

A separate low noise amplifier for the high frequency part would indeed make sense. It would make the circuit more complicated and may need some trimming  (e.g. to get a smooth transition).

[David Hess]

The high bandwidth of the max4239 is of limited use because of the high noise - though there are more noisy ones. At higher frequencies the max4239 is not that bad for an AZ OP with supposedly some 30 nV/sqrt(Hz). Still a scope input amplifier may be lower noise for the higher frequencies, or at least comparable.

May?  MAY?  A scope input amplifier better have a lot lower noise, which is why I often laugh when marketing advertises a modern oscilloscope as being "low noise".

The ones I am familiar with are "chopper stabilized".  "Chopper amplifiers" are a completely different thing.  There are, or were, some other input offset voltage correcting technologies which had no switching noise at all.  I am not sure about "automatic zero".

[SilverSolder]

How much bandwidth does the uCurrent actually need?

Its response seems well behaved and reasonable for its typical use cases.

[David Hess]

How much bandwidth does the uCurrent actually need?

It needs to be fast enough for a multimeter's AC input ranges, which themselves are not all that flat above low frequencies.

[SilverSolder]

How much bandwidth does the uCurrent actually need?

It needs to be fast enough for a multimeter's AC input ranges, which themselves are not all that flat above low frequencies.

I guess another use case is hooking the uCurrent up to a scope, and watching the current consumption unfold on the screen...  but 300KHz seems more than enough for that purpose too, in practice most devices have at least one bypass capacitor somewhere on their rail(s) that limits current pulse bandwidth...

[thmjpr]

Good comment from youtube:

John Wettroth
1 day ago
Hi Dave, I retired as the managing director of product definition for standard products at Maxim a few years ago.  My group and I defined thousands of products like this for over 20 years and I had a hand in the 4239 parts as well as many, many others.   There were a couple issues that came up in this video that I can shed a bit of light on from the chip company side.   Although Philbrick made precsion amps in 50's with tubes,   Intersil , the spiritual father of Maxim invented this category in CMOS chopper amps in the 70's with the ICL7652.  Maxim's second sourced these parts and made improved second sources early in its life.  Here are a few points that may give you a better handle on how chip companies approach high performance analog.  Because of the way that these parts work, the nominal offset is zero.  What gets in the way of this is real third order effect like thermocouple action in the leads and package stress.  Maxim is somewhat unique in that it can and does do post package trim (zener zapping)  vs. just laser trim a the wafer level.  Parts that are perfect at wafer end up being a couple of uV post package due to stresses.  The other thing about very high spec parts like this (100 nV) is testing.  In order to get good yields, test max limits are generally set higher, we called this the threshold of pain.  The typicial values and histograms are there to give you a feel for what you're really going to get.  In difficult to test specs, you're balancing how many seconds you're spending on a million dollar tester- the cost get significant.  Some specs llike leakage currents in analog switches will often have a max spec of 10 uA though will generally be in the femto amp range.  We would release a different external part number that guaranteed a spec like this.  You can get an idea of the real spread by asking a vendor to make you 10,000 (a common minimum) with some spec tightly tested and they don't balk, it means the parts will yield if the customer is willing to pay the delta in test cost (plus NRE)..  Maxim routinely does this for big customers with precision requirements like a test equipment company.  The histograms in the data sheet is taken from the first few wafer lots as the test guys and the design guys dial things in- this process is called correlation.  On a little amplifier parts on big wafers, there be 5000 die per wafer so you can generate a lot of test data pretty quickly.  The voltage range/dynamic range issue is driven by modern processes.  Maxim has a lot of boutique processes but most analog part in the market are made on kind of vanilla cmos processes which are generally digital and low voltage.  These are generally 100-200 micron processes.  If you see cmos parts with +-5V or +-15v supplies, they are fabbed on old 2 micron plus processes or a specialty analogy process.  Maxim, TI and Analog and a very few others keep these old processes running for precision analog.  The economics of doing 130 nm analog on 300 mm wafers makes parts really cheap even though it takes some real design chops to do it.  Your comments about automotive are somewhat on target, they can be robust but mainly this designation has to do to with consistency of supply, change notices and paper.  Sometimes specs are relaxed to improve yields also.  Automotive guys can work with anything but they don't like surprises.  Hope this helps a little.

So explains why other companies aren't near these specs, who knows maybe they even have patents.

One possible alternative, that would be relevant for hobbyist audience: test in production that the Vos is within "tweaking range" then have a note for the user to tweak it themselves on arrival.

[Whales]

Wow, thankyou for finding that.  Re-quoted with line-breaks:

John Wettroth

Hi Dave,

I retired as the managing director of product definition for standard products at Maxim a few years ago.  My group and I defined thousands of products like this for over 20 years and I had a hand in the 4239 parts as well as many, many others.   There were a couple issues that came up in this video that I can shed a bit of light on from the chip company side.   

Although Philbrick made precsion amps in 50's with tubes,   Intersil , the spiritual father of Maxim invented this category in CMOS chopper amps in the 70's with the ICL7652.  Maxim's second sourced these parts and made improved second sources early in its life. 

Here are a few points that may give you a better handle on how chip companies approach high performance analog. 

Because of the way that these parts work, the nominal offset is zero.  What gets in the way of this is real third order effect like thermocouple action in the leads and package stress.  Maxim is somewhat unique in that it can and does do post package trim (zener zapping)  vs. just laser trim a the wafer level.  Parts that are perfect at wafer end up being a couple of uV post package due to stresses. 

The other thing about very high spec parts like this (100 nV) is testing.  In order to get good yields, test max limits are generally set higher, we called this the threshold of pain.  The typicial values and histograms are there to give you a feel for what you're really going to get.  In difficult to test specs, you're balancing how many seconds you're spending on a million dollar tester- the cost get significant.  Some specs llike leakage currents in analog switches will often have a max spec of 10 uA though will generally be in the femto amp range.  We would release a different external part number that guaranteed a spec like this. 

You can get an idea of the real spread by asking a vendor to make you 10,000 (a common minimum) with some spec tightly tested and they don't balk, it means the parts will yield if the customer is willing to pay the delta in test cost (plus NRE)..  Maxim routinely does this for big customers with precision requirements like a test equipment company. 

The histograms in the data sheet is taken from the first few wafer lots as the test guys and the design guys dial things in- this process is called correlation.  On a little amplifier parts on big wafers, there be 5000 die per wafer so you can generate a lot of test data pretty quickly. 

The voltage range/dynamic range issue is driven by modern processes.  Maxim has a lot of boutique processes but most analog part in the market are made on kind of vanilla cmos processes which are generally digital and low voltage.  These are generally 100-200 micron processes.  If you see cmos parts with +-5V or +-15v supplies, they are fabbed on old 2 micron plus processes or a specialty analogy process.  Maxim, TI and Analog and a very few others keep these old processes running for precision analog.  The economics of doing 130 nm analog on 300 mm wafers makes parts really cheap even though it takes some real design chops to do it.

Your comments about automotive are somewhat on target, they can be robust but mainly this designation has to do to with consistency of supply, change notices and paper.  Sometimes specs are relaxed to improve yields also.  Automotive guys can work with anything but they don't like surprises. 

Hope this helps a little.

So the offsets mostly come from package stress and thermocouple action in the leads.  I wonder how the stress contributes -- changes in the thermocouple potentials or unbalancing resistances further into the wafer?

It's also really interesting to read about the voltage limits.  This  explains why many low-power opamps don't go above 5.5-6V, even though it would be really useful if they could go just a bit above 4 AA batteries (6.4V).

[David Hess]

So the offsets mostly come from package stress and thermocouple action in the leads.  I wonder how the stress contributes -- changes in the thermocouple potentials or unbalancing resistances further into the wafer?

I almost discussed this in my post above but decided it was too far from the topic.

Precision analog ICs use a special encapsulation and packaging to avoid stress on the IC die which would ruin precision.  (1) With this and good design, offset voltages below 25 microvolts and offset voltage drifts below 100nV/C are feasible even on a non-chopper bipolar parts.  Chopper stabilized amplifiers do slightly better than this and have the additional advantage of lower, and flat, flicker noise which tends to be an intractable problem.

But in both cases, thermocouple effects are a larger source of error, even for precision bipolar parts.  So special attention has to be paid to layout and circuit design to take advantage of the potential precision available.  Linear Technology application note 9 discusses this and shows measurements of exactly how serious this issue is.

(1) I heard an interview with ... I think it was Bob Dobkin of National Semiconductor ... who told a story about someone in management complaining about the cost of the special encapsulation which analog ICs required and asking why they could not use the same packaging as digital ICs.  Some time after that, someone in management moved the analog IC packaging to the digital packaging plant and a couple months later, the engineers found out what had happened when they tracked down why the yields for analog ICs had crashed.  National lost months worth of production and had to buy parts from competitors and remark them for sale.

It's also really interesting to read about the voltage limits.  This  explains why many low-power opamps don't go above 5.5-6V, even though it would be really useful if they could go just a bit above 4 AA batteries (6.4V).

One of the reasons chopper stabilized operational amplifiers are so common on low voltage CMOS *digital* processes is that there is no alternative on these processes for precision parts.  Even Microchip Technologies makes them.  And being made on a digital process makes them very inexpensive.  But in the case of chopper stabilized amplifiers, you also get their disadvantages like high broadband noise, charge injection, and horrible overload recovery.

[GreggD]

Dave,
KOA Speer TLR2BDTD10L0F75
Res Metal Plate 1206 0.01 Ohm 1% 0.5W(1/2W) ±75ppm/C Pad SMD Automotive

±75ppm/C is not ±15ppm/C but 0.28 vs $4 for Vishay Y14870R01000B9R

Digikey $0.284/1@1000 us$ stock
Mouser  $0.284/1@1000 stock

Searched using octopart dot com

[exe]

±75ppm/C is not ±15ppm/C but 0.28 vs $4 for Vishay Y14870R01000B9R

Vishay is 0.1% part. I think the whole point of using precise parts was to avoid calibration.

[schmitt trigger]

The art of reading and understanding data sheets is a skill which separates the better engineers from the just good.
But I understand that this is not a task everyone enjoys, and I've heard complaints many times that datasheet hunting is "tedious".

I also follow a similar approach to Dave's. But the very first box I tick before starting any search is the "In Stock".
It is frustrating to find your dream part only to realize that it is in the unobtanium status.

[Kleinstein]

Searching only for in stock parts helps to get more common parts. However one may also miss some parts as sometimes even common parts run out of stock. If available have the usually stocked or similar tag helps.

One problem with the search function at Digikey / Mouser is that they sometimes mix typical and maximum values.
The definition of what the typical TC is also seems to be different. Some manufacturers seem to be a little more optimistic than others.  I would have expected something like the RMS value of the TCs found. Over time the process may change (usually improve) a little.

The idea with the µCurrent is to have precision parts and not adjustment, thus the high accuracy parts and OP with lowest possible offset.
For a high BW use the noise could become really important.
For high performance / lowest noise the resistor steps are quite large. At the low end of the ranges the noise gets increasingly problematic. However more resistors would increase the costs quite a bit with the high precession parts.

[SilverSolder]

Searching only for in stock parts helps to get more common parts. However one may also miss some parts as sometimes even common parts run out of stock. If available have the usually stocked or similar tag helps.

One problem with the search function at Digikey / Mouser is that they sometimes mix typical and maximum values.
The definition of what the typical TC is also seems to be different. Some manufacturers seem to be a little more optimistic than others.  I would have expected something like the RMS value of the TCs found. Over time the process may change (usually improve) a little.

The idea with the µCurrent is to have precision parts and not adjustment, thus the high accuracy parts and OP with lowest possible offset.
For a high BW use the noise could become really important.
For high performance / lowest noise the resistor steps are quite large. At the low end of the ranges the noise gets increasingly problematic. However more resistors would increase the costs quite a bit with the high precession parts.

High dynamic range is also important for the µCurrent.  Not sure if there is something clever that could be done here, for example a logarithmic mode?  Is there such a thing as a log autozero amp...

[exe]

Is there such a thing as a log autozero amp...

I don't think so. I was looking hard for an accurate log amp, found none. They were also expensive.

[David Hess]

High dynamic range is also important for the µCurrent.  Not sure if there is something clever that could be done here, for example a logarithmic mode?  Is there such a thing as a log autozero amp...

Dynamic range on the order of 26 bits with reasonable accuracy is feasible with a logarithmic design.  I do not remember ever seeing an automatic "zero" logarithmic converter but it could be done using a pair or more of reference currents.

Since the output would be a logarithm, I am not sure it makes sense without a dedicated meter to show units and I have seen current meters which did exact this to show picoamps to milliamps without range switching.

[Howardlong]

High dynamic range is also important for the µCurrent.  Not sure if there is something clever that could be done here, for example a logarithmic mode?  Is there such a thing as a log autozero amp...

Dynamic range on the order of 26 bits with reasonable accuracy is feasible with a logarithmic design.  I do not remember ever seeing an automatic "zero" logarithmic converter but it could be done using a pair or more of reference currents.

Since the output would be a logarithm, I am not sure it makes sense without a dedicated meter to show units and I have seen current meters which did exact this to show picoamps to milliamps without range switching.

A few months ago I built a design around TI’s LOG114 plus a programmable power supply. On soldered breadboard (Busboard with solid copper groundplane) I could reasonably resolve from 100pA to 5mA, so not at all far off 26 bits.

The problem with this approach is that it’s dependent on a virtual ground, and the device loses its accuracy beyond a few mA as the virtual ground is too weak.

I used an on board MCU and LCD display, and powered it with a battery. I used the MCU’s on chip 12 bit ADC.

You have to be very careful when measuring, I put the entire measuring device and DUT into a foil tray.

[SilverSolder]

High dynamic range is also important for the µCurrent.  Not sure if there is something clever that could be done here, for example a logarithmic mode?  Is there such a thing as a log autozero amp...

Dynamic range on the order of 26 bits with reasonable accuracy is feasible with a logarithmic design.  I do not remember ever seeing an automatic "zero" logarithmic converter but it could be done using a pair or more of reference currents.

Since the output would be a logarithm, I am not sure it makes sense without a dedicated meter to show units and I have seen current meters which did exact this to show picoamps to milliamps without range switching.

The scale could be some number of millivolts per DbV, or something like that, so it would work with any multimeter or scope?