Okapi programmer

Let's start with a picture. It'll save me another thousand words. Click on the picture to get a bigger one. Here comes the Okapi:

The programmer has it all:

Schmitt triggered buffered inputs
MISO, MOSI, CLK and RESET
Programmer auto detect
With decent dual power supply
Programmer powers the target, on demand
Full galvanic isolation between PC and Real World
PC side powered by DC - DC converter; no extra transformer required
Very fast and low power 6N137 opto couplers
PC side power controlled by LM2936 LDO
LED's signalling state of Okapi
10 pin standard ISP connector
Power to target system controlled by reed relais

Above you see the routed board. Contrary to the days of yore, I now mainly produce double sided boards. The price difference is too smal; double sided is perhaps 10 cents more expensive than single sided.
The board features the following:

FR4 material, 35 um copper
All signals have WIDE tracks, at least 40 mils
FULL separation between PC side and real world side
Power/ground planes for both sides
minimum 16 mils of isolation between copperplanes and signals
200 mils of creep space between circuits
Solder masks on both sides
Silkscreening on topside
Four 3 mm mounting holes
Will accept 9 Volts AC or DC

But while routing the board, I came to my senses. Okapi-2 is nice. It beats the competition in hardware, but it is overdone. Why use high speed telecom couplers where we could use cheaper opto couplers that perform better under these circumstances? And why is there no feedback mechanism here? This will be a programmer for programmers and engineers. So feedback is highly appreciated.
So Okapi 2 was aborted. Like so many other projects on this website: I do not only pay attention to succesful projects. The failures also deserve a place on internet.

Also this programmer has it all:

Schmitt triggered buffered inputs
MISO, MOSI, CLK and RESET
Programmer auto detect
With decent dual power supply
Programmer powers the target, on demand
Full galvanic isolation between PC and Real World
PC side powered by DC - DC converter; no extra transformer required
Fast enough, low power, PC817 opto couplers
PC side power controlled by cheap 7805
LED's signalling state of Okapi and programmer interface
10 pin standard ISP connector
Power to target system controlled by opto coupled HEXFET

Above you see the routed board. As usual, I have made a double sided board. It features:

FR4 material, 35 um copper
All signals have WIDE tracks, at least 40 mils
FULL separation between PC side and real world side
Six seperate power/ground planes on both sides
minimum 24 mils of isolation between copperplanes and signals
200 mils of creep space between circuits
Solder masks on both sides
Silkscreening on topside
Four 3 mm mounting holes
Will accept 7.5 Volts AC or 9 Volts DC
Cheaper, yet technically more interesting, opto couplers
2 power indicators and 5 ICSP signal indicators
Open source hardware

To the right, you see the populated printed circuit board for Okapi 3. The basic functions work; i.e. I can make the LED's blink and the power supplies are OK.

Due to an error in the Eagle film maker script (at least in my Linux version) the silkscreen data was not incorporated so the PCB does not look as flashy as I intended. Still, it looks mighty fine without it.
On the boardx is a plethora of LED's for signalling:

PC sided power
AVR target has power
RELAY
MOSI
RESET
CLK
MISO

The latter five LED's are rectangular LED's grouped together in one cluster. They indicate the logic level of the corresponding ISP signal. If MOSI is logic "1", the LED is ON. If the RELAY LED is OFF and the blue LED is ON, then the AVR target is selfpowered.

OK, apparently you are too smart for me. Yes, there's a red circle in the picture and red always indicates something nasty. I made an error. Not a real one, but still real enough to fix it in this stage.

Above you see part of the Okapi 3 ciruit drawing. I use a set of pull down resistors to force the inputs of the HC243 buffer to a defined state if nothing drives the Okapi. This works for three of the four signals. But not for the most important one: the relay that switches power to the AVR target...

When making a circuit one has to take care that the circuit does as little as possible, when idle. So, without a PC or other driver connected, the system should not be active in any way. This is rule one for a safe design.

For some reason I overlooked this, probably due to a glitch in my thinking. The pull down forces a logic '0' on the output B1 (pin 11) of the 243. This makes the opto coupler LED active and hence the opto coupler transistor is conducting. So it feeds +9 Volts to the gate of the HexFET so it opens up to supply +5 Volts to the target system (and the blue LED starts to emit).

This is wrong. In the safe mode, the HexFET should NOT be conducting. So now is the time to come up with a fix that is technically possible without major overhaul of the PCB. I changed it into the circuit below:

This is how it ought to have been in the first place. Too much thinking spoiled it, I guess. Now all four opto coupler input stages are controlled in a similar way. I tried it ands it works. Also, this little fix can be repaired on the ready PCB's with only minor rework. A sharp hobbyknife (or a small electric drill with a ball shaped mill) will do the trick.

On the left you see how the topside is treated. Pin 1 of the first opto coupler used to be connected to the "+5V" signal. Now it's isolated, by milling away the copper and some of the epoxy, just to make sure the cut is through.
On the right you see where to make the cut on the bottomside. Make the cut in the 'channel area' for reasons that become clear in the next stage.

That's it. With these two cuts, plus some fancy soldering in the next phase, we have fixed the design error without having to route miles of wirewrap wire along the bottomside of the PCB. If you have a PCB without the milling, you will need to do this yourself. All PCB's that are shipped from March 21 will be corrected by me.

Below you see the fixes I made on the already built board. On the left the milling (I had to touch the IC socket..). On the right is the fix I made with some pieces of scrap wire from a resistor. I used pieces of scrap metal. But in new boards you can just solder the resistor in pace and bend the remaining wire ends in the right shapes, before cutting them.

Now it's also clear why the cut needs to be in the channeled area of the trace. Originally I made the cut close to the opto coupler pin. With the result that the correction pin came too close to a now exposed piece of copper. So I made a second cut, now in a better place.

Want one? Then get one!

Want one?
Then get one!