Microchip isn't really a player in ADC innovation, I would start looking at vendors like Analogue Devices or Maxim Semiconductor. Microchip has a reputation for skimpy ADC specifications.
That's true - its almost always possible to use better specified, but more expensive parts, but that misses the point. Engineering is about trading off device costs, performance, development costs etc. to meet requirements. Hundreds of thousands and probably many millions of these devices must be sold every year so the engineers who design them in must have (mostly) made a judgment as to whether its accuracy was acceptable, though sometimes they'll get it wrong - such is the nature of dealing in uncertainties.
My question is how would you go about creating an error budget when given a datasheet like this? Its not an uncommon situation when dealing with low cost parts for cost sensitive products. Clearly you wouldn't use it for something that demands five-nines guarantees but it would probably be perfectly ok for, for example, toys where relatively high failure rates at the limits of temperature range and manufacturing tolerances is acceptable. But even there your boss is likely to want to know what percentage are likely to get sent back by customers.
Perhaps my biggest misgivings relate to the seemingly common mismatches between typical figures shown in the electrical characteristics and the typical performance charts. Sometimes you can work out they are simple mistakes and sometimes due to differences in test conditions but often I am unable to reconcile the disparities. At that point you wonder if the figures are remotely realistic especially given that the manufacture won't be bound by them as they are only 'typical'. So do you largely ignore the datasheet and measure as many devices as you can reasonably do and preferably from different batches? And then what - how do you decide how big to make your guard-bands not being privy to the manufacturer's test data and statistics?
If you have enough clout with the supplier then you probably demand the data you need but small businesses generally can't. It clearly helps to have some experience/knowledge of the underlying factors - eg. fet gate leakage may vary over orders of magnitude between individual parts, whereas PN junction voltage/temperature characteristics are fairly well defined by the underlying physics. But, for example, with devices like this ADC its unlikely that you are going to know how the voltage reference is implemented - buried zener or bandgap so can't guess as to its performance.
For what little its worth, I note that these ADCs are used in Chinese 5 digit meters like this:
http://www.ebay.com/itm/Red-0-36-LED-5-Digit-DC-0-33-000V-Digital-Voltmeter-Voltage-Meter-Car-Panel-M42-/301724351834?hash=item46402c3d5a:g:zg8AAOSwF1dUNijsThese are typically spec'd at .3% + 2 counts, 25ppm/C. Chinese specs are notorious and those probably account for the input divider resistors alone let alone the ADC. However its quite possible they do meet these specs in practice - I've got a few and they seem to perform very well (within .03% of my 34401A at room temperature) but are probably calibrated at room temperature. Unfortunately most engineers don't have the luxury of being able to invent specifications.
Apologies if this has gone on a bit but I'm sure there must be many designers here that have had this problem so what do you do? How often is it a case of a wing and a prayer?
Home
Help
Search
Login
Register
