# Freescale v. too many others

One of my pet peeves is when a datasheet lists 'typical' values with no min or max. I’ve written about this a couple of times: **here** and **here** . It seems like this problem is getting worse as components get more complex. In fact, one FAE admitted that “typical” specs are often driven by marketing’s needs, not engineering analysis.

Sometimes the results are almost laughable. TI, which generally does a good job of characterizing their devices, has a step-down converter (the TPS82740A) whose Iq is rated at 360 nA typ, 2300 max. That’s a huge range of values. One can only conclude that the typ number is meaningless. (Not to knock the part – this is a very innovative converter. It will stop switching and connect Vin to Vout through a FET if Vin is roughly at the desired output level. Very cool.)

Maximum values are especially rare in the domain of ultra-low power MCUs. Vendors are in a real slugfest to prove their parts have lower sleep currents than the competitors, and too often “typ” is the only rating given for this critical parameter. Here’s a “typical” example from a vendor of an ARM MCU who makes a big production about their low-current specs:

The careful designer is left scratching his head, with no idea what sort of results he’ll see in a real-world application.

Hats off to Freescale. Perusing the **datasheet for their KL02 Cortex M0+ MCU** I came across the following statement:

**“The maximum values stated in the following table represent characterized results equivalent to the mean plus three times the standard deviation (mean + 3 sigma).”**

Wow! A year or so ago I asked several semi-vendors what ‘typical’ means and none could define it. Here’s a concise mathematical model that makes a lot of sense. Three sigma means you can be sure 99.7% of the parts will not exceed the listed values. 'Typical' still is not defined, but cautious engineers really only care about max values.

A portion of the table that the three sigma rating applies to follows:

It’s interesting that typ and max are generally pretty close together. If typ is the mean – which is only a guess – then the standard deviation they experience is pretty tiny. They have extremely good control of their manufacturing process.

Other members of their Kinetis family are well characterized as well. Some, like the KL03, sip very gently from the power supply in deep sleep modes.

It’s tempting to advertise the best possible numbers for important specs, and ‘typical’ results are a lot more compelling than worst-case values. But these are aspirations, not guarantees, and we engineers can’t design to some marketing person’s fantasy. I hope more companies follow suit.

**Jack G. Ganssle** is a lecturer and consultant on embedded developmentissues. He conducts seminars on embedded systems and helps companieswith their embedded challenges, and works as an expert witness onembedded issues. Contact him at . His website is.

“Freescale's clear definition of the Max value is helpful, but wouldn't it be great if everyone just published the Mean and Standard Deviation? That would allow us to derive our own Max values based upon requirements (e.g. Mean +2 sigma, Mean +3 sigma, Mea

“It's actually a really hard thing to determine mean and std deviation, and if they change, people will sue you.nnRather publish as little as possible and expose yourself to the least possible legal blow-back.nnHow much of this really matters? Very few

“”It's actually a really hard thing to determine mean and std deviation, and if they change, people will sue you.”nnIt is my understanding that the goal in IC manufacturing is to use Statistical Process Control (SPC) to continuously monitor key paramet

“I guess I feel some sympathy for the semiconductor venders. How is it possible to specify what current a device consumes when there are numerous and some times mutually exclusive on board peripherals each of which will have a tolerance and are configured

“Most semiconductor vendors tussle with specification tables, both in the meaning of the parameters and the implied come-back the customer has if a particular part exceeds some limit. It's even harder in the analog world.nnWhen I'm doing a margining exe

“Freescale is defining the maximum well. Hopefully the same is true for the typical. If so the typical would be the mean. Sigma can then be calculated. Of course the mean and sigma will vary over time. “

“As we move to advanced geometries, 6-sigma leakage is off the charts and driving the big deviation between typical and max. MCU power is dominated by cores, speed paths, and the clock tree typically. Peripheral consumption is usually negligible regardless

“Automotive AEC-Q100 parts are apparently treated differently. A document on NXP's website states the following:nn”Six sigma design philosophy is applied to all Q100 devices. This ensures that an end user application designed to the datasheet limits can

“I would very much like to see an article or a video in which a representative of a chip manufacturer explains how MCUs are characterized (e.g. what is the definition of Max) and how they are tested during manufacturing.nnFor example, how many parameters

“We don't automatically get rights as customers.nnWe don't have rights any more than the manufactures do. Where is their right to demand we pay $20 per micro?nnThese are just free market forces in action. We (as an industry) continuously buy parts and

“If manufacturers are working out typical values from a sample of parts then they can (by implication) also give the standard deviations of those values for the sample or min and max values as defined by some arbitrary multiple of sigma around the mean. Th

“I cannot speak for all of Freescale, but in the automotive division, no parts are intentionally shipped that are outside the min/max limits after production test. Every data sheet specification is guaranteed in one of several ways – production test, full

“Thanks for providing those helpful insights. I am interested in learning more about the manufacturing process. My goal is to understand how MCU families and subfamilies are related at the manufacturing level. When a new subfamily is released to the market