For novel ideas about building embedded systems (both hardware and firmware), join the 35,000 engineers who subscribe to The Embedded Muse, a free biweekly newsletter. The Muse has no hype and no vendor PR. Click here to subscribe.
Summary: Does a "typical" parameter in a datasheet have any meaning?
I was pouring over a datasheet recently and the sleep current needed by the part was listed as:
Typical (+25C): 0.01 uA Typical (+60C): 0.04 uA Maximum (+85C): 2 uA Maximum (+125C): 10 uA
What does that mean? The range covers three orders of magnitude.
Decades ago one of my mentors taught me to never use a "typical" spec; always design for worst case. That has been good advice which I've only violated in circumstances where some other data was available or a wide distribution of values was acceptable.
TI agrees. Their datasheet for the LPV521 op amp (http://www.ti.com/general/docs/lit/getliterature.tsp?genericPartNumber=lpv521&fileType=pdf) states: "Typical values represent the most likely parametric norm at the time of characterization. Actual typical values may vary over time and will also depend on the application and configuration. The typical values are not tested and are not guaranteed on shipped production material."
So how does one interpret the current consumption data listed earlier? My first reaction is to use the max (+85C) number as that is, presumably, guaranteed. But that's 200X the room-temperature "typical" value, and room temp is, for this application, the expected operating environment. Alas, the 2 uA at 85C value takes this part out of the running.
While there's really not enough data to use for curve fits, an exponential trend line fits pretty darn well, which is hardly surprising given that much of the current is consumed by reverse-biased P-N junctions. But that's mixing typ and max values, a proverbial conflation of apples and oranges.
I have always assumed "typical" represents an average, or the most-likely value in some probability distribution function. But does it?
I contacted tech support of several semiconductor vendors and asked what "typical" means. The responses were hardly enlightening.
Energy Micro was the first to reply, a mere 12 hours after submitting the question. They said: "A typical for Energy Micro's specs is the mean value of the devices measured. So [it's] not really dependent on the standard deviation. Most CMOS characteristics have a normal distribution, so mean and median are pretty much the same thing."
That mirrors my expectation, but since bigger standard distributions means one can expect a wider range of values the mean doesn't tell us much about what to expect when a product is made in volume.
TI weighed in soon after. "'Typical' values in datasheets are not taken from probability functions. These are taken from multi-lot characterization testing with the mode being considered the 'typical' for a give[n] set of characterization parameters (temperature, Vcc, etc)."
That's interesting but not terribly helpful. The mode alone tells us nothing about what kind of values one can expect. And TI, like most vendors, only lists the typ value (no min or max) for some parameters on at least some of their MCUs so we don't get even a sense of the distribution that can be expected.
Microchip responded: "'Typical' is the average of our characterization data." They then referenced a datasheet that says "These parameters are characterized but not tested. Data in 'Typ' column is at 3.0V, 255C unless otherwise stated. These parameters are for design guidance only and are not tested."
Freescale replied with a link to a Kinetis data sheet (http://www.freescale.com/files/microcontrollers/doc/data_sheet/K60P100M100SF2.pdf). Section 3.8 reads: A typical value is a specified value for a technical characteristic that: Lies within the range of values specified by the operating behavior Given the typical manufacturing process, is representative of that characteristic during operation when you meet the typical-value conditions or other specified conditions
Given that typical is "representative of that characteristic" one might assume this means one can expect the part to behave much like the specified typ data. But the example shown on that datasheet just a few lines later rates a current with a min of 10, typ of 70, and max of 130 uA. The standard deviation is enormous. The typ value teaches nothing about the range of values to expect (though, in that example, the min and max tell us everything we need to know; wise engineers will create a design that works properly over that entire range).
Three weeks after I asked NXP replied: "We have not seen a universal definition of `typical', though it has been one of the most commonly used terms in the industry and product. One interpretation would be that `typical'_ represents `material that comes from our production process under standard SPC (statistical process control.'"
Again, this is interesting and useless, at least for practicing design engineers.
Atmel never responded.
The bottom line is that the meaning of a "typical" spec is devoid of engineering value. If a spec isn't guaranteed, it's not a spec, it's an aspiration.
What's your take? How do you use typical specifications? And what do you do when a spec has only a typ, and not min and/or max, value?
Published June 28, 2013