Anelok: one more MCU - maybe an idea for later

Werner Almesberger werner at almesberger.net
Thu Dec 25 17:28:47 UTC 2014


I wrote:
> - two MCUs, even if both are 32-QFN, cost more than a single
>   one in 48-QFN,

Let me put some numbers: the current BOM cost (1) of Anelok is
about USD 20.30 for 100 units. (2) Today's Digi-Key prices for 100
MCUs are:

Chip	Package	Flash	Product code	Unit price (USD)
-------	-------	-----	------------	----------------
KL26	48-QFN	128 kB	MKL26Z128VFT4	2.795
KL26	32-QFN	128 kB	MKL26Z128VFM4	2.652
KL27	32-QFN	256 kB	MKL27Z256VFM4	3.185
KL16	32-QFN	128 kB	MKL16Z128VFM4	2.314

We also need another crystal (3), in the range 8-32 MHz, preferably
small (the 32 MHz crystal I currently have for RF is 3.2 x 2.5 mm).
Here's one that looks nice and is cheap:

Murata XRCGB32M000F2P00R0, 32 MHz +/- 20 ppm, 6 pF, 2.0 x 1.6 mm,
USD 0.399 @ 100 units.

So we get ...

Chips(s)				Total Flash	Sum of costs (USD)
-------------------------------------	-----------	------------------
MKL26Z128VFT4				128 kB		2.795
MKL26Z128VFM4 + MKL16Z128VFM4 + XTAL	256 kB		5.365
MKL27Z256VFM4 + MKL16Z128VFM4 + XTAL	384 kB		5.898

To this, add a few cents for extra capacitors and maybe a resistor
or two. Regarding total cost, we then get ...

MCUs		BOM cost (1)	Comparison
---------------	------------	----------
KL26 alone	20.30		100%
KL26 + KL16	22.90		113%
KL27 + KL16	23.45		116%

Not too bad.

(1) Components only. To obtain total manufacturing costs, add PCB,
    case, yield, factory fees, labour, shipping, taxes, etc.

(2) For some parts I don't have volume discounts above 100 units.
    For, say, 10k units, I would expect the component cost to drop
    to about half the 100 units figure.

(3) That is, unless we do some crystal sharing. One option would be
    to let the cMCU run from a crystal-governed clock provided by
    the sMCU, but that would have the accuracy issue discussed in
    [1].

    Another option would be to coopt the RF MCU and make it share
    its own (precise) crystal clock. This would of course mean that
    one couldn't use USB during rfkill. But in this scenario,
    rfkill would at least not affect timekeeping by the sMCU.

    One could also mix and match: have a cMCU clock line that
    connects to RF MCU and sMCU and pick whatever we like. I'd
    begin to get a bit worried about EMI, with high-speed clock
    traces crawling all around the board, though.

    Last but not least, if picking a KL27 instead of a KL26 for the
    cMCU, one could use the former's improved RC oscillator for USB
    device mode. (4) One would still have to find a solution for
    USB host, though.

    But for the sake of simplicity let's just assume that the cMCU
    will have its own crystal.

(4) You may wonder whether we shouldn't use the KL27 instead of the
    KL26 even in a single-MCU scenario. The improved oscillator
    would indeed be nice to have and the KL27 comes with 256 kB of
    Flash while the KL26 maxes out at 128 kB. Alas, Freescale
    dropped capacitive sensing in the KL27 (there's still the "GPIO
    method" but I don't know how well that would work), so it would
    be less suitable for Anelok in this regard than a KL26. Maybe
    they'll some day make a KL28 with the best of KL26 and KL27 ...

[1] http://lists.en.qi-hardware.com/pipermail/discussion/2014-July/010676.html

- Werner



More information about the discussion mailing list


interactive