Topic: Technical issue on the Chumby HW

Dear Chumbian -

The i.MX21 processor requires 3.0V nominal (2.7-3.3) for NVDD and you are driving it at 3.3V. This is the same as the Freescale-supplied Cogent board and the ADS, but it is still not right given a reasonable design margin on the power supply. Has this issue been recognized?

I think this is a great product concept and as metro WiFi becomes pervasive (I live in Portland OR), the possibilities will only get more interesting.


Re: Technical issue on the Chumby HW

Hi there,

That's very astute of you to notice it. I had gone back and forth on that point a bit, and it's particularly unusual that Freescale puts the "recommended range" at 1.7-3.3V for NVCC, yet sets the "max rating" at 3.3V. It's pretty darn silly to set your failure point at the same point as your recommended operating point. Note that on QVCC it is 1.65-1.8V recommend, 2.1V maximum--a much more reasonable margin.

I used to design chips on the Freescale Hip7 process that I believe this processor is fabricated in. I've seen the reliability data that their engineers use to derive their maximum tolerances, and their internal methodologies. Honestly, I think their internal models are flawed and I have some disagreements with their methodologies. They come from the Motorola 'six-sigma' lineage, which is great, but it leads to way over-conservative design practices that lead to inefficient implementations. Believe me, I'm *very* cynical now about Freescale's process methodologies, they are just weirdos compared to other foundries I have worked with. When it comes to reliability margins for their dielectrics, I think they tend to underrate compared to other foundries.

I've done some studies myself on their silicon process and my personal theory is that this odd "zero margin" spec comes from an internal conflict in Freescale between a design engineer saying "I think your spec is bogus and this is totally 3.3V compatible" and process engineer saying "according to my precious reliability model these transistors won't work above 3.3V and the sky will fall and babies will cry if you don't listen to me". The design engineer sits down with the process engineer and they have it out. Even if the design engineer was right, and the process engineer was wrong, the process engineer wouldn't budge. Process engineers have a lot of power at Freescale, I find in that organization they tend to be alarmist, and they like their job security. To be fair, process engineers have a very tough job and they have to meet some very stringent specs, but there's being careful, and there's being looney. So the compromise was this odd "zero-margin" spec, where basically a customer is invited to design for a 3.3V type system, but it's ultimately the customer's fault if it breaks, because Freescale can always point to their "zero margin" point and cast a reasonable doubt on the customer's ability to actually meet a zero margin. So essentially, I think it's a wink-wink nudge-nudge yeah it's 3.3V compatible in reality but there's some guy inside the company who disagrees and I can't get around him to get the company to back this up--and if it fails, it's your fault.

While it would be nice to design a system around a 3.0V spec--if you try to design the system around 3.0V, then DRAM specs, and all the other peripheral specs barf--they usually have a range of 3.0V-3.6V, and you run out of margin on the low-end.

If you put the processor on a 3.0V rail, and the DRAM on a 3.3V rail, well, then you have a problem with input overshoot now being 0.3V higher than normal relative to the NVCC rail and powering up the ESD protection diodes on the input pads, which is probably just worse overall for everything. So you can't do that. And I'm really not going to put a level shifter on every data line just for this paranoid spec.

Since all the *other* components in the system *really* are designed around 3.3V, and I think Freescale is just being weird, I decided that it's best to use a 3.3V regulator and then control the tolerance on it as tightly as I reasonably can, or to try and center the regulator at 3.15V, for example, if I really wanted to get nit-picky.

If you read Freescale spec sheets for other processors very carefully, you will find similar weirdnesses all over the place.

7BAA 2E53 01C1 DCFF 497B  E7F0 9699 A303 78F0 D9B9

Re: Technical issue on the Chumby HW

jmore wrote:

I think this is a great product concept and as metro WiFi becomes pervasive (I live in Portland OR), the possibilities will only get more interesting.

I've just started a thread in the "Using Chumby" forum to promote discussion of the possible uses of chumby in metro WiFi zones...I'd love to hear more about what you think they are.