A miniature SMD module as a
replacement for blown uPD5101 or MCM5101 low-power RAM
Here's a homebuilt SMD module to replace blown CMOS RAM in the JRC memory unit NDH-515 or NDH-518.
|© PROJECT ENGINEER:
WALTER - PE1ABR.
Click for larger image
Problem statement and the subsequent dissolving design stages.
Got - a gift!: JRC memory unit NDH-518, blown after a lightning strike and spark-over at the JRC unit.
The spark mark spot (on a screw) is still visible.
On the inside no visible trace damage on the board, but a lot of semiconductors are "exit". (What about the receiver damage......) In particular the (ancient) memory IC's. Yet sin to demolish, repair perhaps? Designing an interface for modern and inexpensive IC's with 32,768 channels? A trial version is already working! But is a little "overkill"!
All TTL and CMOS is cheap and easy to replace, a little harder is the old memory IC MCM5101 or uPD5101 in super low power version (with -LL in the serial number).
The 5101 is really antique, with a total of only 1024 bits in 256-nibble slices of 4 bits. And also with separate input and output for the databits! 256 million bits per chip is actually even antique also nowadays.
A sheer lucky coincidence: on the radio fleamarket in Rosmalen (NL), I ran across a cheap tube of uPD43257BGU-70LL in SMD, in effect to at least 250x greater memory capacity, a complete different kind of CMOS static RAM chip. But with "-LL", superlowpower, suitable for battery-backup! And maybe applicable in an appropriate interface circuit in a small PCB-module with a 22 pin "IC plug". But everything will have to stay super compact to fit!
It would be nice having the entire interfacing also on this tiny board, so that (almost) nothing has to be changed or "cut" at the original JRC board. It shouldn't become much larger in length than the original IC, in the width direction some extra "stretch" is possible.
After some thought-noise a thought has been put on paper. Would it work? How much current would it consume? Will the data I/O interfacing work at Vbatt, or must it due of it's current use be separated and connected to the standard +5V Vcc? To separate is still better, so an extra pin to Vcc is needed unfortunately. PDF link First (error) version. To ease "wiring" on the tiny board: all data and address bits are jumbled, it makes no difference. Don't be worried.
It is recommended building a testcircuit and verify it all, preferably with applying the original SMD components. Where do I obtain an appropriate SMD experiment board? Is rare, or not for sale. OK, then I also design that myself. PDF link to a 10 x 16 cm euro SMD experimental board, type PE1ABR!
Another weekend nicely busy with UV exposing, etchings and soldering on a scrap piece of the euro-SMD board.
After another weekend busy with monkeyhair copper wire (<0.1 mm!) and SMD grit: it works. Current consumption is much too high. More than 60 uA. Standby time only half a year with one set of batteries? Can't be OK, there's an IC leak or something is not right there. Address-bit leak: can not be. After examination of the pull-down on the data I/O signals: that's not either a problem. It turns out to be an external /WE pull-up of 47K connected to Vbatt. Probably is a MEM write protect security at on/off, by keeping signal /WE constant high to Vbatt. The pull-up power is leaking away through the buffer interface via the input protection diodes if the buffer is powerless and Vbatt is present. In short: directly to /WE NO standard inverter transistor nor a HC(T) gate with protection-diodes (in forward conduction) will work, only with real isolated gate inverters or reverse connected diode protection buffers are applicable. Well we use a stage with two BSN type FET's, an isolated gate buffer to connect signals to the data interface. Again a modified sketch: PDF link schematic version 2, the improved version.
Test... It doesn't work? There is absolutely no current use?? But it does work!
Well, well, it works so well now (with super low power consumption in "sleep" mode) that I even doubt whether the IC is connected to the power supply. Less than 1 uA use, least significant bit on the digital current indicator flashes only! 1 uA resolution is the lowest of my meter. Vbatt and Vcc connected together is now no longer possible, as always 1 of the two inverter FET's draws current (> 200 uA).
Testing in trial-mode is successful now, the lay-out work can start on the real PCB. Would It be possible to get it all on the tiny board? Quickly enjoyed many hours with good old-Layo1 for DOS. I like to have an old W98-special occasion up and running with a simple VESA video card, so nostalgic DOS VESA Video programming commands work OK. And it looks fine and works perfectly, and I can handle it. In order to continue using the old Layo1 I have more reasons. The key: the output files generated by Layo1 themselves in postscript (only this version) are of a very clear and easy to understand structure. I can break-in in those files later to get beautiful conversions out of it! I.E. with my own conversion software I can generate and convert from the graystairs output from the Layo1 outputdriver to color and also "clean" the postscript so I become flawless production sheets and perfect documentation in color without strange errors when converting the postscript to PDF format.
A quick SMD PCB view??
Also you can use my software to convert the standard output of layer 1 and 2 not only to a cleaned postscript but also adjusted to the EPS file format. Later you can then merge (by hand) several small postscript jobs in Notepad+ and still use only 1 printsheet, or make macros in postscript to make multiple (identical) print jobs on the same sheet.
PDF link of 10 mini-memory IC PCB-boards in 2 handprogrammed sheets (1 sheet per layer).
Pleasant economical and enormous useful for the same small PCB (again as in this case) when you need a large number of them. After generating a PDF from the "cleaned" postscript you see how it looks, and is subsequently also easier to print from Acrobat on a different (modern) PC with all modern printer options. With the tiny memory PCB and etching them at home the chance to create more waste is unfortunately bigger.
|Now we build them, in many steps and many pictures!|
|Create a whole stack of PCB's, which ensures some reserve... Inspect very thorough for copper short-circuit hairs. And especially when a trace goes between two SMD strips. Ohms testing and possibly correct with a surgical knife!
Use fresh etching solvent. I'm using myself perchloric acid i.s.n. = 10% HCl + 3% H2O2 35% , this is unstable and degrades, it can explode a bottle!!
Before every new use add half a cup of H2O2 35%.
Also use small PVC side-pipes to avoid scratching at the bottom side.
Trace interruptions are worthless, the traces are already pure monkeys hair, how do you manage to glue a repair wire to it??
Take in advance somewhat less booze, shaking hands is a terror to the tiny thungsten drills, because: Drill the VIA (through connecting holes) NOT greater than 0.5 mm. Otherwise the tiny connecting wire-pieces will drown in the solder or even wash away. The rest (header holes) up to 0.8 mm. There is already very little copper on this tiny PCB. And upRIGHT / Vertical drilling, otherwise you're on the other side adjacent to the hole.
Take a piece of flexible power cord, 10 to 15 cm is enough. Add to 1 wire at 1 side a blob of solder, then continue to peel away the rest of the plastic. You now have a bunch of monkeyshair-wire for top-bottom jumpers.
Stretch top and bottom and secure, do not solder one by one, but a few at the same time, a few is 5 to 8 wires simultaneously.
ATTENTION: cut next to the copper trace, NEVER over it.
Avoid large solder bumps under the IC positions. Caused by the selfmade VIA's you already have to bend the SMD legs slightly.
The SMD IC's. Put the PCB in the top edges of a small drill clamp (or vice). Start with the HCT chip and fixate a corner. ALL pins in the right position? Fixate another pin diagonally. Still everything OK?
Gently prick with a needle at the legs to see if all the pins touch the PCB and the SMD chip is not lifted up by touching to a bulge underneath. OK?? Then everything is fixed in place. Use a super-miniature SMD soldering device and do NOT heat too long, the cut connecting through wire-pieces can still flush away. And you cannot correct it anymore afterwards! Then place SMD IC number two.
Now place the former back to the front in the vice. Now the loose SMD gravel is next. Cap's, R's and two FET's, type BSN20 or equiv. (like 2N7002). For the two FET R's use anything between 4K7 and 22K, for the pull-down data anything between 10K and 47K is OK. Use what is available. Floating inputs of unused data I/O shouldn't be done really. A too low pull-down R gives an undesirable loading before there's a real zero written in. So a compromise. The two Capacitors: everything between the 220nF and 47nF is OK.
When merging the print layers together be careful what you securing first, otherwise you can not recover later. Assemble / stack everything first "removeable" (without soldering) to see what is top and bottom of the connecting PCB. For the 22-pin header plug you can use a standard 24-pin (larger) plug and modify, one pin off per strip and remove the middle pieces. Put the two separate strips then temporarily in a standard 22-pin socket to keep the right pitch size and keep stable upright.
The 22-pin connector section is soldered both top and bottom, that's firmer (if you push them in!!) and works easier. Push the strips not so far in the holes that the plastic touches the board! Solder first the protrude tiny points at one side, then solder between the plastic strips and pcb once more all the points.
The pins that are slightly thinner should serve as plug pins!
Again: first stack the second pcb "removeable" to see if everything fits and nothing is upside-down!
Then first solder the assembled interconnecting strips on the SMD IC side.
Apart again and solder again to the bottom-inside of the GND and the 2x +5V points.
Other pins not needed to do.
Then back together again and fix the bottom side, soldering only at the bottom "outside" of the module.
It's rather late, but test a last time for disaster shorts. Then test!
The MEM unit should be without power and also at least 1 backup battery removed. Insert a new memory module. I assume that the memory unit is now meanwhile equipped with professional 22 pin-sockets!
Also an extra +5V connection pole should be made on the main PCB, see further down at "last comments".
Provide a current measuring adapter (very thin double-sided PCB film (photo link) in the battery circuit. Place the last battery. You should see a power glitch and then the current drops to almost zero in micro amps. All my 6 modules together connected to approximately 3.3 V (low voltage NiCd battery used as a low-voltage (worn-out) test battery instead) use only 1 uA! TOTAL! Incredibly low!
(perfect circuit, eh!, is also because the data buffer is on the external +5V Vcc)
And there they are, all nice 6 together!
Finally after 5 years back a working memory unit!
How do you know whether the I/O works OK, and all is well with the data bits?
Are all address lines all well connected?
|Fill channel 1 to 10 of bank A with:
Choose Bank A (all switches adresbit B, C and D would be dis-engaged,
contact problems with those adresbit pull-down than not applicable)
With the table left you quickly see whether there are any address or data bit errors.
Because never programmed channels can contain invalid data noise, you can see the strangest characters or dark spots on the display prior to a real channel MEM write action. This is normal and NOT worrying!
So do: → tune in something → press MEM → press Preset → identical? → OK.
You can quickly erase all noise everywhere: set tune to 000.0, hold MEM, and rotate 24 steps quickly, everything is reset to 0. Next bank.
After a long standstill, the switches B, C and D may have contact cracking, use intensly or perhaps a tiny drop contact cleaner or spray?
I myself had still a module problem.
One display digit did not (always) came back with what I had written in with the MEM button, shit / scheisse.
Power off, battery loose and 2 modules exchanged places. On again.
Problem moved, so it's the module and not a defective LS126 buffer (which were previously all blown too ....)
Seen from front: The rightmost display digit is the leftmost memory IC.
By watching which figure changes in which other figure with the above test, you know what bit it is that doesn't work. In my case an 8 became a 0, and a 9 a 1. Highest bit 4 doesn't want to become a "1".
It is also measured fast and quickly. What goes in the HCT244, what comes out during a press on MEM (write mode). Is it stable the same thereafter (read mode), and what comes out of the HCT244. Memory IC appeares not to accept a "1"!
Although.... at the junction of external D4 (15 + 16) I see a "1", but the corresponding pin 16 of the IC (known there as D5) is not "1". Found. IC is not broken, SMD solder didn't "fluxed" correctly, it is fixed with resin flux instead of solder.
Just a hot tip there → ready!
Please do not accidentaly "shoot out" during the measurements.
Some last few comments
Beforehand in the memory unit an electrolyte must be replaced. At the leftmost IC. This is a "radial" type and is unfortunately in the way. Replacing with a new standard radial 100 uF - 16 to 20 V type, but now lying flat is everything. You might have seen it utmost left on the memory picture above.
If on the data lines the decouple disk C's are a bit large, it is better to move them away a little by pushing them aside.
A few millimeters next to the electrolyte is also an additional +5 V pole for the external +5 V Vcc for the I/O buffer for the bufferstage of the modules. With only one wire to it, which is "looped" from one module to the other.
As a tiny connectorplug a modified jumper plug is used. Works fine!
When this module is being used for another application: pin 18, the /OE of the original uPD5101 is not used in the NDH-JRC units. I haven't used it either in the new circuit. This /OE pin is also NOT available in the uPD43257 IC, therefore you need again another version, the uPD43256, it has only 1 /CE (instead of 2) and 1 /OE pin. But in that case the second CE2 is not available.
|Original scanned old NEC Datasheet uPD5101L|
|Original NEC Datasheet uPD43257BGU-70LL|
|The 10 x 16 cm euro SMD experimental board, type PE1ABR!|
|The final interconnect and interface schematic sketch (version 2), the improved version|
|The documentation version of the tiny SMD pcb|
|The 10 mini-memory IC PCB-boards in 2 handprogrammed postscript sheets (1 sheet per layer)|
|And finally a link to the original unmodified circuit diagram of the NDH-518|
| To the Photographs section of the NRD-515 page
by Walter - PE1ABR - 2012-03-22