PCBs | George Smart

While tinkering with a homebrew GPSDO project, I spent a bit of time searching the depths of the internet for information and parts for my project. I found a PCB for a Symmetricom GPSDO (specifically the Symmetricom 089-03861-02 as per the PCB silkscreen in my case, although the firmware reports itself as 090-03861-03). The board was cheap because the OCXO was missing – I suspect it had aged beyond where the error voltage range was specified, making it unusable. But the board was around £15 GBP delivered from AliExpress, so I took a chance on it based on the fact it had a Furuno GT-8031F GPS receiver which is a GPS module specifically designed for timing applications which “delivers highly accurate GPS timing”. The module also had a CPU (Renesas F2317VTE25V-H8S/2317) with accompanying flash and RAM ICs as well as a Xilinx Spartan-3 FPGA (XC3S200).

Besides the missing OCXO module, my board worked perfectly. I was able to piece most of the information together from the following two resources:

I saw that some people had put their units into Hammond Manufacturing project boxes, and that gave me a few ideas. It would be nice to box the unit up with some ancillary electronics to share the UART (57600 baud, 8N1) over Ethernet TCP/IP. However, my experience of UART to Ethernet modules has always been poor and friends reported similar, so I opted for an FTDI-based USB interface (namely the FT232RL, as the cheapest/easiest part during the chip shortage of 2019-2022). I was tempted to use a Maxim MAX232 device to perform the necessary conversions, but, in the end opted for two NPN transistors – a bit hacky, but much cheaper. I may well come back to the Ethernet option in the future.

The carrier takes 5V at around 2A input on a 2.1mm x 5.5mm DC power socket, a USB-B connector and holds five LEDs: four main LEDs from the GPSDO (which are connected to test-points on the PCB, I can’t find the LED signals on any connector), plus one LED from the FTDI device showing USB activity.

The image below shows where the LED signals are taken from. For reference, the big IC in the centre is the Xilinx Spartan FPGA:

The LED signals are all common anode, fed from +3.3V taken from the board. The LEDs are fed through a resistor, and the individual cathodes connect back to either the CPU or FPGA through another of the wires:

RED: +3.3V supply used to power the LED anodes
BROWN: Alarm
WHITE: Activity
YELLOW: Heartbeat (DS2)
ORANGE: Error (DS1)

There are two other LED signals (DS3 and DS4) which aren’t connected.

From here, I created a carrier board which had the correct mounting holes for the Symmetricom module, and offered front LEDs to show status, an easy 5V interface and a USB-UART interface to the module control port. The final board looks as below:

With the module fitted, the board looks more like the following. Note, this was an earlier version of the PCB:

The project was designed to fit inside a Hammond Manufacturing 1455N1601BK case which takes a 160mm long by 100mm wide PCB, and the designed PCB fits the case nicely. By default, the Hammond case comes with either aluminum or plastic end plates, but I went to the effort to make PCB front and rear plates to enclose the case. Using PCB meant I was able to use tools I was familiar with to create the end places, and use the same order as the main carrier PCB. The copper on the PCB could be used to create a metal Faraday screen to enclose any electrical switching noise, and the silkscreen could be use to add legends, logos and decals.

I prototyped the designs as below in the PCB package and then used my laser cutter to test them out for fit. A test fit of the front panel is shown below:

The below picture shows the PCB end boards as they arrived and the cardboard cutouts. There are some small tweaks, between the two, but overall the process worked well. One thing to note is that the internal cutouts for the BNC connectors are a little bit tight – in future I need to add an extra 0.5mm or so clearance (in addition to the 0.5mm clearance I left already). However, the connectors pushed in even if a little tightly.

Final assembly went well, and I am very pleased with the results:

While chatting to a friend recently about an old Z80 CP/M machine I built based on Grant Searle’s Z80 CP/M machine, I decided to fire the machine up for a demonstration and give it a try.

The machine had been in a shelf in my conservatory for well over a year since I last powered it, and when I connected the power, it was clear that there were signs of life, but, the machine was not happy at all! I could see bus conflicts on the data bus as well as very ‘broken’ behaviour.

After probing around, I decided to check a couple of logic gate (those controlling the address and read/write lines) using my TL866ii+ programmer which conveniently supports logic gate testing and RAM testing. All of the logic gate ICs confirmed to be working correctly, as did the 128Kx8 RAM chip. Checking the EPROM yield a different story!

Forgetful EPROM

Some of the bits towards the end of the ERPOM had been erased (with UV erased parts, this means a ‘0’ bit had become a ‘1’ bit). The snapshot below shows in red bytes that had failed the verification.

The first 0x2012 bytes were completely unaffected with 324 differences in total (just under 2% of the 0x3FFF bytes in the 16KB ROM) mainly concentrated in the end of the address range.

This teaches me to always cover the UV window with something opaque as per the datasheet’s advice. I know people say that sunlight can erase such devices, I never really believed them. As an aside, others have looked into this, which I had not seen until now: see this hack-a-day article for a timelapse of EPROM data when exposed to sunlight.

And, secondly, it left me in a quandary as to how I should restore the ROM code.

Initially I thought I should just be able to reprogram the device over the top of the existing code, since the changed bits would be reprogrammed back to ‘0”s from their current ‘1’ state. But, this turned out not to work. I’m not sure why, but the programmer would fail at 50% – just about the time the corrupt bits arrived.

I decided the next thing to do was to try and erase the IC – but how!?

Erasing UV EPROMs

I didn’t have one of the classic EPROM eraser boxes with a timer and a nice antistatic mat. I’d have to improvise. Essentially tinker around until the PROM was erased (reading 0xFF in all locations).

Eprom Eraser UV141 Professional made by Industrial electronics Ltd. UK

My first attempt was with a DSLR camera flash. I had seen these cheap EPROM erasers online that look like a xenon flash tube, so figured it was worth a shot. However it made no difference at all – I assume the camera flash has an UV filter.

My second attempt was with a UV LED, inspired Charles Ouweland‘s investigations along similar lines. Charles reports that this took 48 hours to erase the PROM. Like always, I was in a rush, so this didn’t really work for me. I could have driven a roundtrip to my parents (Dad has a proper UV eraser) in less then 4 hours including a cup of tea and a chat with Mum!

The third attempt was to use a UV insect-o-cutor which features UV fluorescent tubes as in the UV141 eraser pictured above. I placed the chip on the desk mat and put the UV insect-o-cutor directly on top such that the finger guard was on the top of the IC and the bars were not obscuring the window.

The scary part of this was that the high voltage aspect of the bug-zapper was popping on occasion with small insects – thankfully, the insects popping did not obscure the IC window nor cause ESD damage! After 20 minutes (the usual time I run the UV141 for) the IC was exactly the same. No additional bits had be cleared to logic ‘1’s. I suspect the output is UV-A and not the required UV-C.

Making a UV-C EPROM Eraser

I went back to the second approach with the LED, this time consulting the EPROM datasheet (extract below), and noted that the recommended wavelength for erasure is 2537Å (253.7nm). This puts the UV light required to erase the EPROM in the UV-C spectrum.

I set about trying to find an LED with the correct wavelength. Some more digging showed up the VLMU35CB20-275-120 UV-C (270-280nm peak) LED offering a typical 13.5mW. Such wavelengths are common for curing acrylic nails and for sterilising surfaces in medical applications. At the time of writing, one such LED costs around £4.50 with VAT.

Constant LED Current

These LEDs appear to require between 5.0 and 7.5V at 120mA. I opted to create a simple constant current device using an LM317 set to 120mA and the two LEDs in series. This should work fine when supplied with at least 18V (7.5V + 7.5V + 3V [LM317 dropout voltage]). A SOIC8 LM317 is capable of 200mA and low profile enough to allow the LEDs to be close to the EPROM without catching. R1 is chosen to cause a voltage drop of 1.25V (the LM317 reference voltage) at the desired current. See this TI flashback for more details.

R1 = 1.25V/I = 1.25/120mA = 1.25/120e-3 = 10.4Ω. The closest easily available value is 10Ω, which checking backwards would give a current of 125mA. Since the datasheet mentions 150mA supply giving increased power, I have chosen to go with 10Ω as this will not damage the LED.

We should also consider the size of R1. The power dissipated in the resistor is easily calculated:

P = I^2 x R = (125mA)^2 x 10Ω = 120e-3^2 x 10 = 0.144W = 144mW.

So, a standard ¼-Watt through-hole resistor would be fine. For surface mount, 0805 resistors are typically 100mW, so a combination of multiple resistors should be used to handle the power. This could be four 10Ω parts in series-parallel, two 22Ω resistors in parallel (making 11Ω [114mA in the LEDs]), or two 4.7Ω resistors in parallel (making 9.4Ω [133mA in the LEDs]). I’ll opt for two 4.7Ω resistors since this is still below the maximum recommended working current of 150mA.

Designing the PCB

Initially I had considered tacking two wires on the LED and using a lab power supply to power the LEDs. However, I figured if I was making a constant current supply – mainly to avoid a “oops!” moment and pop £8 worth of LEDs – I would make a PCB to house it all. This way, I could keep the board with the EPROM programmer in the programmer box. The design brief would be simple. The PCB should fit inside the EPROM programmer (TL866ii+) box. The PCB should be (maximally) 110x50mm. I figured 100x50mm would be a nice size. A 5.5×2.1mm DC jack would allow for easy connection to wall-wart PSU or other power source. The rest was just two LEDs, and LM317, two resistors and a decoupling capacitor or two!

Since the PCB is quite simple, I figured it would be a good candidate to make with a laser cutter. Above shows an overlay of the Gerber edges and the layer to be etched. The back areas are there copper is to be removed. The white areas will remain copper.

The board was created by coating a standard single-sided FR1 copper PCB blank with black paint from an aerosol can. This will be used as the etchant mask. The laser cutter is then able to remove the mask by burning away the paint layer, exposing the copper to the etchant.

Typically, I’d also cut a solder mask from mica sheet to help apply paste, but for such a simple board, it isn’t worth the extra effort and waste materials.

The next job is to etch the PCB in ferric chloride – you should strive to make a much better job than I did!

I completely over-etched by board by having the etchant too cold and leaving it far too long! But hey, it’s been 10 years since I last made a PCB in my (parents) kitchen sink!

From the image above you can see that my 10 thou track has completely disappeared in the right etch, and has almost completely disappeared in the left. I then proceeded to make a mess of drilling the connector holes, misreading a 4mm drill as a 2 mm. The board ended up as a total mess, but, I got it working!

My apologies for making such a mess of the board. However, as you can see, the two LEDs are lit and a current of 117.3 mA flows through the series LEDs. I may tweak the resistor values to get an extra 10mA or so, but for now, the result is good enough!

So, now to erase the EPROM

With the PCB made up, I eagerly put the EPROM back into the reader and loaded the old ROM code in. I reverified it to see that indeed 324 mismatches showed up, just as before. Confirmation that all my previous attempts had failed!

I held the newly made PCB above the UV window and gave the EPROM a 1 minute ‘blast’. I read the EPROM again, curious to see how many extra bytes failed verification. Too my surprise, 1088 byte failed verification – considerably more than before. I looked through the data and could see a pattern; the EPROM was blank. Sure enough, the 1 minute blast was enough to clear the EPROM!