Keithley 2281S-20-6 can't measure Ah for more than 10 hours accurately

[tom66]

Here's what happens if you try to use a Keithley 2281S-20-6 (battery emulator) to measure the total Ah charge consumed by a fixed load - in this case a 33k ohm resistor across a 3.3V source.

You would expect a nice linear plot as constant current = linear slope of accumulated charge.  And you get that... for the first 3 hours or so.  Then the slope changes.  And it changes once more about 3 hours later.  Finally, it levels out at around zero by about 8 hours into the test.

These points vary considerably with the load current: they are not consistent.

We have been wasting our time trying to track down why the energy consumption with our product varies over time, we would expect the consumption to be steady once it has settled down.  2.5 months wasted chasing this bug only to discover it is the instrument introducing the bug.

Keithley/Tek have not been particularly helpful although to their credit they are now looking into the issue with a fix anticipated in the next couple of months, and have offered a loaner precision multimeter which might allow us to make the measurement we need.

However for a device to ship - and go through numerous software updates - with an error like this is baffling. Did anyone actually test the Ah integration logic?  It seems that whatever accumulator they use has a terribly small resolution - I wonder if they are just using an IEEE754 32-bit float?  Given the error seems to halve and double it suggests that the resolution of an internal accumulator is the problem here.  Hopefully they don't just change all `float` to `double` - this needs a proper precision fixed point accumulator.

This has really given us a sour taste regarding this instrument.  If you pay £2,500 for a SMU you don't expect your department's time to be wasted tracking down bugs like this but perhaps this is part of the problem with software on T&M kit now: patch it when the customer complains.

[Ice-Tea]

What do you mean you don't expect your department to chase down bugs?!?

I fully expect to chase down bugs, faulty cables, footprint errors, datasheet ommisions, programmer malfunctions and last but not least human stupidity. It's commonly refered to as "job description" 

That being said: holy shitstorm batman.

[HighVoltage]

It would be interesting to find out if this is just your particular unit of all 2281S that have this problem?
Maybe try a real SMU for comparison, like the 2450 or 2460.

[tom66]

Since we're using the Ah accumulation facility, we're not going to be able to compare other Keithley SMUs here.   The current reading is always accurate. But, as the accumulator gets larger, the accuracy of that accumulator reduces.

It would be a fair limitation if it could only accumulate up to, say, 100,000 Ah before reporting a saturation error, but this occurs very early on at a charge of around 300uAh, which is really not much, I mean you'd expect many smartphones to eat through that just flashing up a notification or something like that, yet they advertise this instrument as "perfect" for evaluating such devices.

A 64-bit fixed point accumulator would allow a high 100kAh max accumulation while having a lower bound of around 5 femto-amphours