Vintage TMS2764 EPROMs acting strange... Schrödinger's EPROM?

[SilverSolder]

I took two TMS2764 EPROMs out of a piece of equipment, erased them, and programmed them with the latest firmware (with a MiniPro programmer).

So the weird thing is - it verifies 100% OK in the programmer...  I am able to read the code out and compare on the PC, 100% the same.  But the EPROMs don't work in the actual circuit!?!

I tried with some Intel EPROMS instead and they work fine.   All of the 7 TI EPROMs that I have are behaving the same way as described above...


I am struggling to understand how the programmer can read the EPROMS, but the CPU in the equipment they came from cannot?!?   

[MK14]

If it was a poor quality programmer, e.g. the proper programming voltages were too low, or not enough time was used, to properly program in the bits. I suppose, the programmer casually (relatively slowly) reading/verifying back the bits, might see the correct values.
But a relatively high speed processor (slow, by today's processors), needing rapid from the EPROMS (slow by today's standards, probably) data accesses, might be responding too slowly, because the programming has been done too faintly.

N.B. Above is speculation, there are other possibilities.

Have you covered the EPROM windows ?
Sometimes they can be sensitive to visible light, and stop working properly, without opaque labels on them. Ideal labels, seem to have aluminium (?) strips inside them, for near perfect no light leakage.

[BrokenYugo]

Maybe the burn profile is wrong, check datasheet vs programmer voltages? I've heard of programming them twice to really make sure the gates are charged fully, maybe try that. 

[MK14]

Also, the access times can vary between different chips (datasheets). Are the 'new' ones fast enough ?

[SilverSolder]

The voltages and timing all checked out vs. the datasheet. 

I got them working by programming them multiple times with the same data, thanks for the ideas.

Maybe these devices are just getting old and cranky?  They are 30 - 40 years old, after all...

They seem to behave more like analog than digital devices! 

[MK14]

The voltages and timing all checked out vs. the datasheet. 

I got them working by programming them multiple times with the same data, thanks for the ideas.

Maybe these devices are just getting old and cranky?  They are 30 - 40 years old, after all...

They seem to behave more like analog than digital devices! 

EPROMS can be quirky/fussy, in all sorts of ways.

My understanding, is they are actually ANALOGUE inside (with digital decoding for the addressing and input/outputs, obviously), with op-amps/comparators, performing the actual readings. There are probably internal reference voltages, for logic '0' and '1', and the comparators then select the most appropriate logic level, based on the analogue voltage from the selected EPROM memory cell.
Similar to modern flash memory, but without the electronic 'erase' capabilities. So they use UV light, instead.

[perieanuo]

hi,
it happened to me with AT24C01 eep once before 2000, the programmer said "program ok, verify ok", the eep didn't work in the machine.
well, the programmer was right, the machine was right.
why? cause programming algo for at24c01 (in fact, atmel and xilinx versions of 24c01 are differents from st24c01 mainstream eep's) is different for ST24C01 algo.
you can check datasheets
test with some really serious programmer and we will see after, minipro is for hobby, maybe it' somehow the same problem

[CJay]

I took two TMS2764 EPROMs out of a piece of equipment, erased them, and programmed them with the latest firmware (with a MiniPro programmer).

So the weird thing is - it verifies 100% OK in the programmer...  I am able to read the code out and compare on the PC, 100% the same.  But the EPROMs don't work in the actual circuit!?!

I tried with some Intel EPROMS instead and they work fine.   All of the 7 TI EPROMs that I have are behaving the same way as described above...


I am struggling to understand how the programmer can read the EPROMS, but the CPU in the equipment they came from cannot?!?   

Sounds like they needed a slower programming cycle, I'm curious to know if you tried verifying them again after they'd failed in circuit?

[MK14]

Sounds like they needed a slower programming cycle, I'm curious to know if you tried verifying them again after they'd failed in circuit?

My partly vague recollection, was that there was essentially 2 extremes of programming algorithms for EPROMS.

The original VERY SLOW, but reliable/robust program every memory location, properly. It was something huge like 50mS per location, hence rather/very slow.

Various 'fast' algorithms, that did something on the lines of programming a location for 1mS, and immediately reading it back. If it was correct, moving to the next byte, if not, programming it for another 1mS, until it is correct, then moving on and doing that for all the bytes that need programming.

So, the wrong algorithm (or a poor quality programmer), wouldn't work out too good.

Also, maybe the chips have deteriorated over the decades ?

[MK14]

hi,
it happened to me with AT24C01 eep once before 2000, the programmer said "program ok, verify ok", the eep didn't work in the machine.
well, the programmer was right, the machine was right.
why? cause programming algo for at24c01 (in fact, atmel and xilinx versions of 24c01 are differents from st24c01 mainstream eep's) is different for ST24C01 algo.
you can check datasheets
test with some really serious programmer and we will see after, minipro is for hobby, maybe it' somehow the same problem

A while ago, I was reading an article, about the differences between cheap/hobby programmers, and very expensive professional/production ones. I think it was for PIC chips.

It was something like the professional/production programmer, used an accurate 5.5V to program, then an accurate 4.5V to read the verification data back, or something like that. To ensure the programming would be robust/reliable for 10+ years, and at differing ambient temperatures, etc etc.

So the cheaper hobbyist equipment, can appear to 'work'. But will the device it just programmed, still be working in 10+ years time and/or at min or max allowable operating temperatures, supply voltages, etc.

[CJay]

Sounds like they needed a slower programming cycle, I'm curious to know if you tried verifying them again after they'd failed in circuit?

My partly vague recollection, was that there was essentially 2 extremes of programming algorithms for EPROMS.

The original VERY SLOW, but reliable/robust program every memory location, properly. It was something huge like 50mS per location, hence rather/very slow.

Various 'fast' algorithms, that did something on the lines of programming a location for 1mS, and immediately reading it back. if it was correct, moving to the next byte, if not, programming it for another 1mS, until it is correct, then moving on and doing that for all the bytes that need programming.

So, the wrong algorithm (or a poor quality programmer), wouldn't work out too good.

Also, maybe the chips have deteriorated over the decades ?

From memory some were faster than 1mS as well, they'd hammer the cells with charge until it read correctly then 'overburn' it with a few more for good measure and move on to the next byte, tended to be only lower voltage devices and was generally pretty reliable but I did have a few occasions with second hand chips where it wouldn't be 100% reliable, I rarely ever had a problem with slow programming that I couldn't attribute to a faulty part but some second hand chips (Seem to recall Fujitsu were the worst offenders) would pretend to be erased and then suddenly have remnants of their old contents or just refuse to take new data.
 

[drussell]

They seem to behave more like analog than digital devices! 

Well, yeah...  They are more like analog devices than digital. 

[Gyro]

I do vaguely remember that there was something a bit different about TMS27xx devices. I can't remember what it was -  different Vpp / more programming failures, /OE access time or something, but there was some reason that we used to dodge them when grabbing blank devices from the communal eraser tray. Best to find the specific datasheet and check it very carefully against something like a Hitachi or Fujitsu. Sorry, it really is a very vague memory.

[perieanuo]

hi,
it happened to me with AT24C01 eep once before 2000, the programmer said "program ok, verify ok", the eep didn't work in the machine.
well, the programmer was right, the machine was right.
why? cause programming algo for at24c01 (in fact, atmel and xilinx versions of 24c01 are differents from st24c01 mainstream eep's) is different for ST24C01 algo.
you can check datasheets
test with some really serious programmer and we will see after, minipro is for hobby, maybe it' somehow the same problem

A while ago, I was reading an article, about the differences between cheap/hobby programmers, and very expensive professional/production ones. I think it was for PIC chips.

It was something like the professional/production programmer, used an accurate 5.5V to program, then an accurate 4.5V to read the verification data back, or something like that. To ensure the programming would be robust/reliable for 10+ years, and at differing ambient temperatures, etc etc.

So the cheaper hobbyist equipment, can appear to 'work'. But will the device it just programmed, still be working in 10+ years time and/or at min or max allowable operating temperatures, supply voltages, etc.

I used in the past programmers better than tl866, like wellon one if I recall right, then smartprog ii  and now batego
I recall I had devices bad programmed with wellon, including eproms, then erased and succesful done with smartprog
for me, it's either the programmer either the algorithm, like I said first time.things like tl866 have some flash/eep/eproms in the list they don't really get written with the correct algo specs from memory datasheet imho (but they report verify done)

[SilverSolder]

I took two TMS2764 EPROMs out of a piece of equipment, erased them, and programmed them with the latest firmware (with a MiniPro programmer).

So the weird thing is - it verifies 100% OK in the programmer...  I am able to read the code out and compare on the PC, 100% the same.  But the EPROMs don't work in the actual circuit!?!

I tried with some Intel EPROMS instead and they work fine.   All of the 7 TI EPROMs that I have are behaving the same way as described above...


I am struggling to understand how the programmer can read the EPROMS, but the CPU in the equipment they came from cannot?!?   

Sounds like they needed a slower programming cycle, I'm curious to know if you tried verifying them again after they'd failed in circuit?

Yes, I did try that - they verified 100% directly after failing in circuit!  -  Perhaps the programmer loads the lines less than the (35 year old) circuit?

[SilverSolder]

hi,
it happened to me with AT24C01 eep once before 2000, the programmer said "program ok, verify ok", the eep didn't work in the machine.
well, the programmer was right, the machine was right.
why? cause programming algo for at24c01 (in fact, atmel and xilinx versions of 24c01 are differents from st24c01 mainstream eep's) is different for ST24C01 algo.
you can check datasheets
test with some really serious programmer and we will see after, minipro is for hobby, maybe it' somehow the same problem

A while ago, I was reading an article, about the differences between cheap/hobby programmers, and very expensive professional/production ones. I think it was for PIC chips.

It was something like the professional/production programmer, used an accurate 5.5V to program, then an accurate 4.5V to read the verification data back, or something like that. To ensure the programming would be robust/reliable for 10+ years, and at differing ambient temperatures, etc etc.

So the cheaper hobbyist equipment, can appear to 'work'. But will the device it just programmed, still be working in 10+ years time and/or at min or max allowable operating temperatures, supply voltages, etc.

I used in the past programmers better than tl866, like wellon one if I recall right, then smartprog ii  and now batego
I recall I had devices bad programmed with wellon, including eproms, then erased and succesful done with smartprog
for me, it's either the programmer either the algorithm, like I said first time.things like tl866 have some flash/eep/eproms in the list they don't really get written with the correct algo specs from memory datasheet imho (but they report verify done)

Not only did the programmer (TL866CS - supports 21V) report verify done.   I was able to put one of the programmed chips into it, read it into a file on the PC, and compare that file with the original...   100% match.

So, it seems that perhaps the amount of loading on the data bus is different on a modern programmer made with modern ICs, compared with the loading that the EPROM sees in the actual circuit?  So a "marginal" EPROM works fine when not loaded too much, but fails when some older circuits need more current on the data bus, or some other signal lines?

The programming algo for the EPROM appears to be mostly in line with the data sheet:



Here are the settings chosen by the programmer:




If I try to increase Vdd (write) to 6V as implied by the data sheet, the programming fails immediately with corrupt data, and the chip has to be erased and "done over".  It also fails at 5.5V -  so I am wondering if this data sheet may have been superceded...

In any event, all 6 EPROMs are still working fine in the actual circuits this morning.

I did not have these problems with Intel EPROMs,  I may source some more Intel ones if any of the TMS ones lose their data...

[wraper]

Those EPROMs might be simply faulty/degraded due to their age. There was a thread discussing EPROM outputting some weird reduced high level voltage occasionally.

[SilverSolder]

Those EPROMs might be simply faulty/degraded due to their age. There was a thread discussing EPROM outputting some weird reduced high level voltage occasionally.

Yes, I think they have gone half deaf.  But by shouting at them long enough, they got the message! 

[CaptDon]

I have a hand built programmer and I use the full program time allowed per byte. It takes MUCH longer to program but they will always work and retain their data through the worst conditions.

[wraper]

Also it's not clear what the circuit is. Maybe it needs faster of slower EPROM to work properly.

I took two TMS2764 EPROMs out of a piece of equipment, erased them, and programmed them with the latest firmware (with a MiniPro programmer).

So did it work properly before with the same EPROMs?

[SilverSolder]

Also it's not clear what the circuit is. Maybe it needs faster of slower EPROM to work properly.

It is an 8080 CPU, it runs at a leisurely pace and can add wait states if necessary.  These EPROMS are much faster than it needs, so there will be no waits.

[SilverSolder]

Also it's not clear what the circuit is. Maybe it needs faster of slower EPROM to work properly.

I took two TMS2764 EPROMs out of a piece of equipment, erased them, and programmed them with the latest firmware (with a MiniPro programmer).

So did it work properly before with the same EPROMs?

Yes, exactly.  The recently purchased "new old stock" TMS2764  chips behave exactly the same way as the TMS2764 devices that were in the device originally.

All of them respond to being "shouted at" repeatedly.  I've programmed 6 of them now,  each one repeated 8x over for good measure...  they appear to work reliably after such a therapy session.

[CaptDon]

I did my programming using just one cycle of maximum allowable length per address/byte. I think the old 2716's were 60ms max, apprx 2 minutes to program, and the 2732/2532 max was around 50ms or nearly 4 minutes to fully program. I think I was hitting my 2764 27C64 at around 25ms single pulse. Never killed one and never had it read or function incorrectly. They have been in several systems for over a decade and still running. I never liked the 'Program it just enough and move on', my stuff was way to critical in its mission.
 

[james_s]

I've encountered several situations where a programmer would read an EPROM fine but it wouldn't work in the actual circuit. In one case the ROM in a vintage terminal got corrupted and I was able to read it and then re-write the exact same chip and reinstall and it fixed the problem. Access time can also be an issue and they seem to get slower some time after programming. I once repaired a vintage arcade game board and it looked fine but then after about 30 minutes or so increasing amounts of noise started to show up in the sprite graphics. Eventually I figured out the EPROMs I used were too slow, they would work fine fresh off the programmer but after some time they showed lots of random pixel noise.

[SilverSolder]

I did my programming using just one cycle of maximum allowable length per address/byte. I think the old 2716's were 60ms max, apprx 2 minutes to program, and the 2732/2532 max was around 50ms or nearly 4 minutes to fully program. I think I was hitting my 2764 27C64 at around 25ms single pulse. Never killed one and never had it read or function incorrectly. They have been in several systems for over a decade and still running. I never liked the 'Program it just enough and move on', my stuff was way to critical in its mission.

The max on this particular programmer is 1ms, it doesn't allow any higher setting as far as I can tell.  So, repeated runs is the only way to increase the exposure time with this programmer.  At least it is able to do that, so there is a solution!