Recommendation aksed for embedded processor

I am looking for a processor/controller to be used in an embedded system that offers the following:

- low costs (if possible < 30 $ @ 10k)

- stable and well-supported platform

- PCI integrated (33MHz/32bit, no PCI-X, maybe PCIe could be interesting if it is already available as integrated peripheral)

- Gigabit LAN integrated (option: could be done via PCI with special IC, but then it has to be included in the costs)

- support of linux (preferrably "standard" linux)

- at least 300MHz CPU clock, better more for performance reasons (may depend on actual architecture)

Nice to have, but not crucial:

- USB 2.0 integrated (option: could be done via PCI with special IC, but then it has to be included in the costs)

- hardware accelerated encryption

- hardware accelerated network operations, e.g. for TCP/IP etc.

On first view the Intel network processors looked interesting, but they offer only 100Mb LAN (and most of them exceed the price limit anyway...).

Any hints where to look are appreciated.

Matthias

Reply to
Matthias
Loading thread data ...

Difficult to do for the following reasons.

To support Gigabit, you probably need 64 bits PCI and faster than

300MHz CPU. You also needs lots of RAM buffers, and things gets expensive beyond 64M. What wrong with 100M plus a switch?
Reply to
linnix

Just to clarify: This amount is meant only for the processor, all other things have an extra budget. But still, it is a tough constraint... If there processors matching my wish list but costing something above

30$ then I would like to learn about them, too.

Regarding PCI we have to assume that PCI 33MHz/32bit is sufficient, since we do not need to transmit continously 1 Gbps, but the device will experience periods where it has to transmit say 150-200 Mbps for several minutes. Most of the time the data rate will be even below

100Mbps. On the other hand, if the Gb-LAN interface would be an integrated peripheral of the processor then the (external) PCI bus would not have to carry the network load.

Regarding CPU I agree. 300MHz seems to be the absolute lower limit assuming a powerful processor architecture, but more realistic are >=

500MHz.

Sometimes we need data rates above 100 Mbps directly from the processor.

Any hints where to look are appreciated. Matthias

Reply to
Matthias

What about multiple 100M links ? There are quite a few 32-bit processors with 4x embedded 100M ethernet interfaces. Used in cheap switches AFAIK.

Regards Anton Erasmus

Reply to
Anton Erasmus

I am not sure about the price, but what you describe will be covered by one of the Freescale PPC based devices. Have a look at the 8548 (I believe fastest) or the 5200 (only

10/100 Ethernet, though, but has PCI 32/66). You will probably be able to find some Linux for those, and before too long I'll probably release a fullblown DPS for the 5200.

Dimiter

------------------------------------------------------ Dimiter Popoff Transgalactic Instruments

formatting link

------------------------------------------------------

Reply to
Didi

May I suggest the AMD Geode GX2/CS5536 (533 MHz), for $40 @1K. $30 @10K is possible, if we buy together. We only need a few hundred of them. We are using RTL8100 (100M, 100 QFP), but you can use RTL8169 (1G, 128 QFP).

Works with Linux 2.2 and 2.4, but have not try 2.6 yet. Works with Dos and Window 98 (customer requirement, don't ask why), have not try XP yet.

Yes for PCI.

RTL8169 is possible.

Other features: VGA and LCD, USB 2.0, 64 bits DDR SDRAM, less than 10W total power.

Major drawback: no CPU cache and lower CPU performance.

Reply to
linnix

Thanks; I had a look on the Freescales devices and found some very interesting ones. Although they exceed the original price limit they will be taken into consideration since they offer features that may reduce the total chip count significantly.

Particularly the MPC8349E is attracting me (seems to be around 60$, but I do not have accurate prices yet). Any comments on that one?

Regards, Matthias

Reply to
Matthias

Thanks for your reply. Since we are trying to reduce the number of "complex" chips this approach does not exactly fit. Also, the performance may be an issue, then.

Nevertheless, this information is appreciated as well, since it helps in decision finding.

So it seems that I will have a closer look on the Freescale devices...

Regards, Matthias

Reply to
Matthias

Hi Matthias,

looks good to me. The core is 603e compatible, this gives you the standard FPU etc., and the first DPS I am likely to release for outside-tgi use will be based exactly on it.

Apart from the 1mm pitched BGA which will be a tough route (perhaps impossible on only two signal layers), I see nothing scary.

Dimiter

------------------------------------------------------ Dimiter Popoff Transgalactic Instruments

formatting link

------------------------------------------------------

Reply to
Didi

This cost around the same as the GX2/5536 pair, and similiar complexity.

It's possible, but expensive for 2 signals + 2 planes. You would need

0.4 pad, 0.2 holes and 0.2 spacing. We just got a batch of such boards, with 30% failure (not passing the board testing) at the factory.
Reply to
linnix

How many traces do you manage between the BGA pads? With the 1.27 pitched BGAs, I manage 3 traces and this is enough. At 0.4 pads, do they drill 0.2 through them? I doubt 3 traces will be doable, this would take 3.3 mil (0.086mm) design rules, those I have worked with do not want to hear about anything below 4 mils. I am not sure a big BGA will be routable with only 2 traces between the pads... Please give some more details, I am curious.

Dimiter

------------------------------------------------------ Dimiter Popoff Transgalactic Instruments

formatting link

------------------------------------------------------

Reply to
Didi

2 traces with reduced width between pads.

No, they drilled standard off-pad via, so called dog-bone pad. Although we looked hard at plugged BGA pad, we chickened out at the end. We might try to experiment more before doing that in production. With plugged BGA pad, it might be possible to go below 1mm (more layers, without trace between pads) for smaller holes or bigger holes for lower cost.

For the plugged BGA pad, we made some donut PCB prototype and filled them with high temperature solder. We then layered a low temperature solder balls on top. Finally, we mounted the BGA (more like ball on ball, instead of ball on pad) and reflow with low temperature.

The low temperature solder is expensive (silver) and we need a better alignment process for ball on ball. So, we need more testing before putting them in productions.

I agree. Finer pitch designs are more expensive than more layers.

---------------------------

Reply to
linnix

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.