152 replies to this topic

[Spaced Cowboy]

I'm glad you said that, because I was beginning to wonder if it was sufficiently Atari-focussed to warrant the constant updates

 

I'm actually having a huge amount of fun playing with machining Aluminium for the first time - up until now I'd only tried plastic and wood (which the machine dealt with, with aplomb), so don't worry - updates will keep coming I've not played with metalworking since I was a kid at school, so vague memories of emery paper, brazing and welding are all being dragged out of very cold storage... I'm having a lot of fun though [grin]. The Fab is still in the process of laying down the chips on the PCB anyway, so it's not as though there's anything else I can be doing ...

 

The cage actually looks a lot more impressive when you hold it in your hands compared to the photos, although maybe that's just me, because I know how much effort has gone into making it

 

If I'm being critical, the cage is a bit flexible across the back panel (if you bend the sides away from each other), but I'm pretty certain everything will be locked into place once I get the base-plate on. It's a bit like making some of that furniture from MFI or (to a lesser extent) Ikea. The bookshelf sways alarmingly until you tack down the plywood across the back

[Spaced Cowboy]

Small design update
 
The new design fits onto a single Carvey Workspace, partly because I reduced the slot lengths a bit to better match the PCIe blanking cards, and partly because I shortened the internal brace as well.

The stack-up inside the cage will be 

• Brackets to hold the base-plate, thickness of ~1mm
• The base-plate, thickness of either 1mm (if I use the same Aluminium for the base) or 3.175mm (if I use 1/8" thick Aluminium sheet for the base)
• Spacers to hold the PCB off the metal surface. These will be another 6.35mm (1/4"). I'll probably use metal spacers and transmit ground to the PCB via them
• The PCB itself, which will be 1.6mm thick.

So the top of the PCB will be ~1cm above the bracket supports, and the baseplate ought to have slots which the 8cm long PCIe blanks can slot into. The new design below takes this into account.

 
The top of a PCIe bracket has a hole for the screw to go into, and a dimple to mate with a corresponding hole on the cage. The total overall length of the inner side is 8cm, with 1cm of that being the tab that slots into the base-plate.

The issue here is that we need to be able to use a thumbscrew (preferably) to secure the bracket at the top, but we're screwing that thumbscrew into a piece of Aluminium that is only ~1mm thick. There aren't many threads in 1mm ...

So the plan is to use rivet-nuts and then we can use these (or any other M3 nut) to secure the PCI brackets in place. The rivet nut will give us several mm of screw thread for the thumbscrew to gain purchase on. Using rivet-nuts means that I have to drill a larger hole though - in this case a 5mm diameter hole for a 3mm diameter screw, which is why the holes seem to be offset a bit incorrectly for the PCIe-bracket screw-hole - they're sized for the rivet-nut instead.
 
 

. .
 

[Stephen]

Sorry if I missed this, but what CNC device are you using to cut out these aluminium panels?

[Spaced Cowboy]

Sorry if I missed this, but what CNC device are you using to cut out these aluminium panels?

 

It's a Carvey from Inventables, although if I were buying anew, I would probably go for an X-carve instead (from the same company). Larger work-area, more oomph in the spindle, and easier to hack. For an intro to CNC though, you can't beat the Carvey, it makes things pretty darn simple - even the design interface (which I now just import .SVG files into then move them around) is web-based and very slick.

 

I can't say enough nice things about Inventables - the after-sales support is amazing. The products are excellent at reasonable prices, what's not to like ?

 

Having said all that, I showed the backplate to the lab-manager at work, and he asked me "why didn't you cut it on our laser cutter ? Then promised to set me up with an intro to the machine, so hopefully I'll be able to get laser-cut pieces soon, which ought to be a lot faster (it takes about 5 hours cumulative time to CNC the cage). The Carvey elides about 0.1mm of Aluminium per pass... There are a *lot* of passes...

 

Simon

[Stephen]

 

It's a Carvey from Inventables, although if I were buying anew, I would probably go for an X-carve instead (from the same company). Larger work-area, more oomph in the spindle, and easier to hack. For an intro to CNC though, you can't beat the Carvey, it makes things pretty darn simple - even the design interface (which I now just import .SVG files into then move them around) is web-based and very slick.

 

I can't say enough nice things about Inventables - the after-sales support is amazing. The products are excellent at reasonable prices, what's not to like ?

 

Having said all that, I showed the backplate to the lab-manager at work, and he asked me "why didn't you cut it on our laser cutter ? Then promised to set me up with an intro to the machine, so hopefully I'll be able to get laser-cut pieces soon, which ought to be a lot faster (it takes about 5 hours cumulative time to CNC the cage). The Carvey elides about 0.1mm of Aluminium per pass... There are a *lot* of passes...

 

Simon

Awesome!  I'd kill for access to a laser cutter.  I have some nice metalworking equipment (all Miller, MIG, TIG, oxy-acetylene, and a Plasma cutter) but nothing CNC, so all of my work does have a "hand built" quality to it.

[Spaced Cowboy]

Awesome!  I'd kill for access to a laser cutter.  I have some nice metalworking equipment (all Miller, MIG, TIG, oxy-acetylene, and a Plasma cutter) but nothing CNC, so all of my work does have a "hand built" quality to it.

 
 
Yeah, I spoke to the lab-tech on the way home, and he showed me the Laser Cutter, and told me to calendar a intro-session with him. Thing is, I don't think this'll cut Aluminium - it's an Epilog laser, and they only go up to 150W. I think you need 400W to cut Al. It'll certainly "engrave" Al, but not cut it IIRC.
 
Maybe there's another one, or maybe they've got the super-secret upgrade, I don't know - I'm still going to calendar the session so I'm "approved" to come in after hours and use it. I'm sure there's things I'll find that I can do with it

 

[Update]

Had my "Laser printer 101" course. Nothing too problematic - don't cut anything with Chloride/Flouride/Cyanide in the name... Look new-to-you materials up first, as well as the general mechanics of the process. All well and good.

 

The bad news: It's a 75W CO2 / 50W Fiber laser, ain't no way that's cutting Aluminium. 

 

The good news: Getting the PD account will let me use the (very nice Statasys) 3D printer and 4' x 2' bed CNC machine as well. The CNC has been bought, but is awaiting installation in the lab in the new Campus, which won't be for a couple of months, but this is an order of magnitude nicer machine than anything I'm going to ever buy...

 

So, for the time being, I'm still going to be using the Carvey for pieces (and to be honest, I'll probably continue to do so for prototyping even when the big machine is in-place. It'll be nice to be able to run a batch job of 20 pieces though  

[Stephen]

 

 

Yeah, I spoke to the lab-tech on the way home, and he showed me the Laser Cutter, and told me to calendar a intro-session with him. Thing is, I don't think this'll cut Aluminium - it's an Epilog laser, and they only go up to 150W. I think you need 400W to cut Al. It'll certainly "engrave" Al, but not cut it IIRC.

 

Maybe there's another one, or maybe they've got the super-secret upgrade, I don't know - I'm still going to calendar the session so I'm "approved" to come in after hours and use it. I'm sure there's things I'll find that I can do with it

Hmm - perhaps you can custom engrave me some emblems for my 87 Daytona Shelby Z

[Spaced Cowboy]

So, up until now, when I place a piece of Aluminium onto the Carvey workspace, I look across the edge of the metal, to see which way it wants to bend. Then if the top-face bends up, I'll place it top-face-down, to minimise the deflection in the center - since the metal is only clamped at the outsides. This way, the natural stress is to force the metal into the waste-board which gives a relatively flatter surface on which to carve. It has its own problems when you're taking off the last 0.1mm, because the springy metal then wants to jump upwards as the last of it is released, but placing a sufficient number of tabs generally gets around that issue.
 
Somehow, I got this wrong, last night, and placed it the wrong way up. Drilling holes is ok, because the pressure is vertically downwards, and the drill will force the metal onto the waste-board surface, but when the CNC starts to move sideways to carve out channels, things can go very awry, very quickly...
 

 
This is the start of the first pass. The CNC machine is supposed to cut 0.1mm of metal per pass to protect my bits and stop from overloading the stepper motors. Even the first plunge went further in than I would want, and then it got progressively worse, to the point where the machine was routing through the full 0.8mm of Aluminium sheet. It was also complaining, bitterly, about being forced to do so. I hit the E-stop pretty quickly! Fortunately, the bit is fine, and the machine seems to be no worse for the wear.
 
[aside] 
It doesn't help that (because Aluminium is prone to melting if placed under too much stress), I'm using an up-cut single-flute bit. A "flute" is basically a cutting-edge combined with a spiral escape-chute for the material that's been cut by the cutting-edge. Typically I'd use a 2- or 4-flute bit for wood, because the number of flutes multiplies your productivity - a 4-bit flute will, all things being equal, cut 4x as fast as a 1-flute bit, and so you can dial-up the speed of the machine as it makes passes over the material. However, for my machine, it's important to use an up-cut 1-flute bit for Aluminium...

• The up-cut bit forces the fragments of metal upwards and away as they are removed, this is important because otherwise a tiny fraction of that fragment might melt under a down-cut bit, and then again a little later, etc. etc. The result is that your end-mll bit becomes blunt and eventually useless.
• The single-flute bit is important because Aluminium produces (relatively) larger flakes of material as it is machined. If there are too many flutes, the "escape routes" for the removed Aluminium aren't large enough, they get clogged, and you're back to the problems in the previous point.

On some (read: more expensive than I'm willing to pay for) machines, they can mix it up even more - there are machines that cut underneath oil or water, ones that spray water/oil on to cool the material, some that have air-feeds to help blow away (and cool) the flakes of material removed etc. All this is to do with preventing the cutting edges of the bit from heating up too much, and melting the Aluminium. Remember, the spindle on a CNC machine is rotating really fast - mine (a relatively slow one) runs at ~12000 rpm. There's a lot of energy being dumped into the contact point of bit and material...
[/aside]
 
So there's no new update with pretty pictures today, nor tomorrow either. As I've mentioned, it takes 5 hours or so to perform a carve, and I have an "All hands" comms meeting today from 3:30 to 6pm. There'll be no time for a new carve by the time I get home - my machine is in the "Man-shed"[+] at the bottom of the garden, close by the neighbours house (who don't have so large a garden as us). I like to keep the noise down at night, and carving metal isn't exactly quiet. It's not as loud as I thought it would be (presumably because I'm only taking off such small amounts per pass, *usually*) but it's still too loud for night-time carving.

[+] I used to have a room where all the toys would be put, but that quickly disappeared when the kid came along. These days I'm relegated to the shed. It's a nice shed, though

[Spaced Cowboy]

So this isn't quite the end-result for the cage (I still have to do the base-plate, and there's some small tweaks that I could make to the screw hole positioning), but it's definitely on the home straight...
 
Here's what it looked like just after the cut was complete on the Carvey
 
 

And after taking it off the workspace, popping out the internal bars, putting the rivet-nuts into the top-bar, bending the appropriate pieces in the appropriate ways, it can be assembled thus:
 
 

.

... and from the outside:
 
 

..

As far as design goes (if not colour) it actually blends in quite nicely with the rear of the existing case. Now if only I had black-anodized Aluminium... I *do* need to take off the tabs with a file, and smooth down the edges of the metal to get rid of any burrs, but this is basically it, apart from the base-plate, which isn't hard. I'm going to hold off making the base-plate anyway until the circuit board is nailed down, because the base-plate will need mount-points for the PCB, and the location of those is going to depend on the layout.

Simon.

[Spaced Cowboy]

Economies Of Scale

 

So, with the card-cage pretty much sorted out, and the PCB in the assembly stage at Seeed, it seemed like a good idea to start ordering components. I don't usually use the larger 1206 surface-mount components, but in a gesture towards making the board easier to solder, all the non-automatically-assembled components are pretty easy to solder by hand. Even the trickiest one (the USB mini power-connector) isn't difficult because you only need to solder the two outer connectors on each side - the rest are not connected.

 

With that being the case, I needed to order the parts. I've had 5 boards assembled, so interesting quantities are:

 

• board-quantity-1 : this is what someone who got a PCB with all the chips / difficult components pre-soldered would need to order from Digikey (or obtain somewhere else, of course) in order to solder the remainder of the board. 
• board-quantity-5 : this is what I need right now
• board-quantity-50 : this is what I'll need to order *if* we ever get any significant interest in the project. At this point, we'd be looking at a custom case as well, probably.

 

So here's how it pans out:

 

 

The q50 price has a guesstimate for two of the PCBA components (they're not stock at Seeed); it also doesn't include shipping, but neither of these will significantly alter the price IMHO. Note that this is just the XL-side of things, there'll be another (similarly priced, I hope, depending on what I put on it) board that actually hosts the PCIe slots.

 

Also worth making crystal clear is that the price for the PCBA includes the price of the PCB manufacture, as well as all the chips that are placed on the board, and of course the work to actually place and reflow-solder the chips.

 

Simon

[Spaced Cowboy]

Black is the new Black
 
So I couldn't resist. I ordered some anodised-black aluminium sheet, made the small adjustments to the screw offsets, and told the machine to have at it. Here's the start and end pictures of the milling.
 

.

I did have some issues milling this out, I was trying to conserve the anodised black sheet, and I chose the best orientation for that, but unfortunately this orientation wasn't ideal for keeping the Al sheet clamped tightly down on the waste-board. Because I'm using an up-cut bit (see the aside above) there's a constant up-force on the sheet, and since I'm cutting out a large amount of the sheet, there's less structural integrity left anyway.

The upshot is that at the end of the carve (when it's doing the outside cut to shape the panel from the blank, I was getting a lot of vibration, and the bit started to try to chew through too much aluminium at once. That resulted in an E-stop, and the placement of the painters-tape along the offending edge (where there aren't any clamp-holes to hold the part down - I might rectify that). You can see the effects if you look closely at the horizontal section of the back-panel, but I can live with it for now

Anyway, fortunately I was present when this presented, so I could fix it and carry on. Once the carve was complete, and after separating it, filing off the tabs (which only takes seconds - aluminium is a soft metal), I started to put the rivet-nuts into the top. Four of them went in fine, the fifth was a bit dodgy, and the rivet-nut tool failed on the 6th rivet. Using the tool, you screw on the nut, then put it into place, squeeze the handles and the nut is deformed in place to make a rivet - then you just unscrew the tool. This "heavy duty" tool managed a total of 12 rivets before the screw-thread was stripped, rendering the tool useless. I'm feeling a negative Amazon review coming on...
 
Anyway, here's what it looks like - I'll have to do the two left-most rivets when I get a replacement tool, but the system looks a lot nicer in black aluminium than silver...
 

.

This is with the internal support-plates in-place as well - you can just about see them if you zoom-up the close-up picture, but otherwise the back-panel covers everything nicely. I was slightly concerned that filing the edges of the pre-anodised sheet would leave the edges outlined in silver, but in fact filing vertically downwards prevents that effect.

Simon.

[Dropcheck]

Nice.....

 

How's the electronics coming? 

[Spaced Cowboy]

Nice.....

 

How's the electronics coming? 

 

 

I'm waiting on Seeed at the moment. The board has been with them for a while, so it ought to be coming back soon-ish. At least I hope so. They say an average of 25 business days, and it's been that long as of today, but there was a 4-day holiday in the middle of that, so I may have to wait another week or so.

 

Impatient doesn't begin to describe it On the other hand, we're getting a new addition to the family tomorrow, so mundane things like electronics will take a back seat for a few days, so it's all good

[mytek]

Next board I would suggest trying ALLPCB. They are very inexpensive, good quality, and ship via DHL. So from date of order to arrival at your house you are looking at 5-7 days. And if you are doing larger and/or 4-layer boards, than go with EasyEDA.

[Spaced Cowboy]

Next board I would suggest trying ALLPCB. They are very inexpensive, good quality, and ship via DHL. So from date of order to arrival at your house you are looking at 5-7 days. And if you are doing larger and/or 4-layer boards, than go with EasyEDA.

 

 

In this case, I think it's the 'A' (assembly) in the PCBA process that's causing the issue. Since I wanted a few boards, and I didn't want to do all that fine-pitch soldering, I got them to do all the hard-to-do SMD work, leaving me with just the 1206-sized parts Much easier than 'e=0.4mm' pads...

 

Reading the FAQ, It's supposed to be a 3-day turnaround for assembly once the BOM has been sourced, which I believe it has. The board has been in the "PCBA in production" for 12 business days now, so I've just pinged them. We'll see what they say.

 

I appreciate the pointers though. Of the two, Allpcb seem to do assembly as well, so maybe I'll look at them if seeed turn out to be dragging their feet too much.

[Spaced Cowboy]

Seeed update
 
So, as mentioned above we got a new puppy last week, and most of my time has been taken up with her, but I did ping seeed to see what the delay was on the board. It turns out there were two problems (both my fault) which were holding up the assembly. They sent me the below ...
 

... and pointed out two things:

1) The part I'd specified in the assembly BOM for U$9 (which they had dutifully ordered) was the wrong form-factor. I'd gone to the schematic to find the part, searched at Digikey (which brought up a whole selection of form-factors for the part) and given them SN74CBTLV3257PWR as the manufacturer's designation, rather than SN74CBTLV3257DR. That's a TSOP-16 part as opposed to the actual SOIC-16 form-factor that was required; there's no excuse, I just wasn't paying sufficient attention. If they have any of the SN74CBTLV3257DR in stock (I doubt it), they'll create the board with those parts, but I'm expecting to be shipped boards which don't have that part soldered. I can solder a SOIC-16 part, so it's not the end of the world, but I am annoyed with myself for this one.

2) The second error was that there's no silkscreen for the XMOS chip, so they didn't know where pin-1 was. There is a silkscreen in the Eagle layout...
 

... and I hadn't caught the problem when reviewing the gerbers. Still, that was easy to sort out - I just sent them a modified version of the first image above, indicating which pin was pin-1, and a copy of the second image for good measure.

3) U$10 is highlighted in their image, but apparently they didn't need any help with that after all - I confirmed with them that there was nothing more they needed from me after resolving the above 2 issues.

I would have been happier if they'd reached out to me when they found the problem, rather than waiting for me to get in touch, but all's well that ends well. The boards ought to be shipping to me on Wednesday. Who knows, with the puppy around, I may even find time to do something with them when they turn up [joking]...

Simon

[mytek]

Simon it could of happened to anyone, especially while you are still in the R&D/Prototype stage. I find the whole SMD package thing quite confusing myself. BTW, the board's looking great     

[Spaced Cowboy]

All a-board!
 
So the boards came back from Seeed I took a moment to make sure I was static-neutral w.r.t them, and then opened up the first board. The boards looked really good with all the chips in the right place and orientation, but then my eye caught the area around the XMOS chip and my heart skipped a beat... There were several solder-bridges on the 0.4mm pins...
 

 
The whole point of sending them off for assembly was to get a clean well-soldered placement of this chip in particular. The other ones are much easier because they have larger separation between the pins. A moment later I realized these bridges were probably by intent - it's been so long since I designed the circuit that I'd forgotten the details. Looking at the CAD design...
 

 
there is a 1:1 correlation between the "solder bridge" and the points where the pins are joined together. Typically it's VCCINT or VCCIO, but RST_N and TRST_N are also tied together on the XTAG interface.
 
So I haven't had time to solder up a board to start the bring-up, but here's what one looks like before I start attacking it with a hot air gun...
 

... which looks pretty good to me. Notably absent is U$9 (I guess they didn't have any in-stock, which isn't much of a surprise), so I'll have to order some of the larger form-factor size chips. I'm pretty busy in work for the next two weeks (got a demo to give at the end of next week) so I'm not sure how quickly I'll get to starting bring-up, but I'm pretty pleased with the end-results so far...

Now let's just hope the design was correct

[Dropcheck]

How much did these boards set you back.?

[Spaced Cowboy]

Including the cost of the PCB and components, about $81 per board.

Bear in mind that it’s a 5/5 space/trace 4-layer board, and the components come to almost $40 (I think, I’ll check and update later, I’m just out picking up the evening meal )

It gets cheaper per board if you hit a reasonable volume, of course... I think that’s ok. I also managed to hit their $50 off in March deal, which helped.

[update: it's actually ~$29 for component costs, and $74 for the total cost per board, so I was a bit off with $81 and $40. I still think it's a good deal for the assembly - the PCB at these low volumes is ~$10 so assembly of the 17 chips/crystal is costing me ~$35/board, even though I only had 5 made]

[Spaced Cowboy]

One step forward, two steps back.
 
So I started populating the board, first doing the power-areas, and the 3.3v isolated area worked well.

Problem #1: I noticed that the scheme I'd put together for isolating power at bring-up didn't take into account the fact that the board was already partly populated by the assembly. I need to make the 3.3v power go directly to the monitoring circuit, not just to the general circuit net. There's no way as it stands to check the 1v (which monitors the 3.3v supply and only switches on once the 3.3v rail is alive) without bringing the 3.3v rail generally up.
 
So having seen that 3.3v worked well, I chose to just install everything, then I'd make the 3.3v solder-bridge connect, and then I'd be able to check the 1.1v rail.
 
Problem #2: As soon as I made the solder-bridge to allow the 3.3v to break out onto the rest of the board, the 3.3v supply dropped to about 0.4v, and the LED dimmed dramatically. Apparently, I had a short. I spent some time looking for one - the parts I soldered myself are all pretty large, so I was surprised there was a short at all, and I couldn't find one anywhere.
 
At some point, I decided to check the chips again, and the board looks like:
 

Notice the SDRAM chip above the SD-card. That dot in the upper left corner ought to be in the lower-right. Seeed had put the chip on the wrong way around (on all the boards, to boot). So, after turning the air a virulent shade of blue, I took the chip off using a hot-air tool.

Unfortunately, while removing the chip, I managed to remove some of the pads as well. Blue air intensified... I think I can patch it up (although it'll be tricky) because the pads connect to other visible points on the board, but I'm not best pleased right now.

Once I had the SDRAM chip off the board, I checked the 3.3v and 1v rails again, and they worked fine, so I'm pretty sure that's the problem. Hopefully, if I can patch the five broken pads, I'll be good to go to test further tomorrow.

 

Simon

[Spaced Cowboy]

Striding forward like a boss
 
Ok, much better news now. I managed to patch the 5 pins on the SDRAM on the board, making it the ugliest board with SDRAM I've ever seen...
 

 
with a close-up of the ugliness looking like:
 

 
... yeah...
 
HOWEVER, when I plug in the XTAG adaptor board (connects the Mac to the XMOS cpu via USB) as in the first photo above, I get a green LED (generally a good sign) and if I then scan for devices, I get:

sh-3.2$ xrun -l

Available XMOS Devices
----------------------

  ID	Name			Adapter ID	Devices
  --	----			----------	-------
  0 	XMOS XTAG-3         	3CUoiRh.	O[0]

Which is XMOS's way of declaring an 'O' type device at node 0 in the chain, connected via an XTAG-3 adapter board. By contrast if I unplug the xe-interface board from the XTAG and do the same scan, I get:

sh-3.2$ xrun -l

Available XMOS Devices
----------------------

  ID	Name			Adapter ID	Devices
  --	----			----------	-------
  0 	XMOS XTAG-3         	3CUoiRh.	None

So I now have communication to the on-board XMOS chip at least. Now I can start exploring the various peripherals, installing a simple LED flasher onto the XMOS chip (the hardware equivalent of the standard software 'hello world!' application) and then I can try plugging it into the 1088XEL and see if I can source power correctly from the bus.
 
So, much better news than last night

[mytek]

Wow that must have been fun  .

 

In the future you might want to get a spool of 'wire wrap' wire to have on hand for this kind of patching. It's got a much tougher and thinner insulation which can not only take the heat better, but is a very small wire gauge which is better for doing this tiny rework.

[Spaced Cowboy]

Wow that must have been fun  .
 
In the future you might want to get a spool of 'wire wrap' wire to have on hand for this kind of patching. It's got a much tougher and thinner insulation which can not only take the heat better, but is a very small wire gauge which is better for doing this tiny rework.

 

Not as much as you might think No, really!

 

Good call on the wire-wrap wire though. The heat-sensitivity of the PVC cladding caused me to throw away a few promising connections once one end was soldered. It was immensely frustrating.

 

At work, there is a lady named Janet who is an absolute wizard at this sort of thing. My cunning plan is to offer to buy her lunch for as often as she demands, if she'll do the remaining 4 boards. Not being a klutz like myself, I'm sure she'll be able to remove the SDRAM chips from the board without killing the traces. She regularly does 0201 work (parts that are 1mm by 2mm in size). Yeah, that's 36x smaller than the 1206 ones I have on the board...

 

One other thing I forgot to mention is that resistance tolerances almost caused me some grief as well. The 1.0v line is pretty tightly specced, and it's created by the ratio of two resistors from a variable-output supply. The actual output was coming out as 1.3v, which is way too high - 1.1v is the 'absolute maximum' specified in the data sheet.

 

Fortunately I have a fair amount of chip-scale resistors in 1206 form-factor, and I could replace the 5.1k I'd originally soldered in (which was really 4.7k according to the Fluke multimeter) with a 3.9k one (which was really 4.2k). That was sufficient to bring down the voltage to 1.05v, which is in spec  For any future revision of the board, I think I'll go with a fixed-voltage output device rather than rolling my own. It'll be slightly more expensive but could save a lot of grief...

 

Anyway, as it turns out, the isolation of the power-supplies was a good thing after all

[Spaced Cowboy]

So I'm not going to post here for every little thing that I get working, but this is a milestone, so it deserves a post

 

I got the XMOS workspace set up, with the correct (24MHz) oscillator defined, and the chip configured to run at 500 MHz. This was actually pretty easy once someone on the xcore forum pointed out I'd missed a checkbox on the 'New Project' dialogue

 

Getting a trivial 'xc' program up and running  ...

#include <xs1.h>
#include <timer.h>

port leds = XS1_PORT_4F;

#define NUM_PATTERNS 6
int patterns[NUM_PATTERNS] = {1,2,4,8,4,2};

int main(void)
    {
    int delay = 50;                         // delay 50 ms
    int counter = 0;                        // A counter for the patterns array
    while(1)
        {
        leds <: patterns[counter];          // Drive the next led pattern
        delay_milliseconds(delay);          // Wait

        counter++;                          // Pick the next pattern
        if (counter == NUM_PATTERNS)        // If we are at the last pattern
            counter = 0;                    // then wrap around.

        }
    return 0;
    }

... gives me the traditional hello-world hardware equivalent of "blinkenlights" up and running (warning: movie is ~10MB). 

 

If you look on the video, you can see the little yellow switch close to the XTAG connector. That switch allows power for the board to be sourced from the XTAG itself, so there's no need to power the board separately, which is nice

 

In other news, Seeed sent me an email asking if the board was all ok... So I sent an email back to them with a photo showing how the SDRAM was rotated. Not sure what they'll do, but thought it was nice of them to follow up.