DIMM testing: Is there any solution?

[Jane1]

What hardware test should I use to find out a faulty memory chip in DIMM module?
My understanding is that if any RAM chip ( used in DIMM ) is faulty then the DIMM is considered ( by the system that uses that DIMM) to be faulty .
I can replace ALL RAM chips on the DIMM but it is not what I want to . Also a replacement of DIMM is not what I want.
Does anyone have any experience with such repairs or where I can learn more?
.
Thank you

 

[betwixt]

Actually finding out which chip on the DIMM is faulty is fairly easy as it shows up as a data error at a particular address or range of addresses and that is normally tracable to one device. The commercial DIMM testers I have used are big and very expensive machines with matched cable impedances and cable lengths to the DIMM contacts to ensure there is no skew between signal timings so unless you have deep pockets I would forget using one.

The simplest system I have seen was ingeniously simple, it used counters to scan the row and column addresses (selected by RAS and CAS signals on the IC) and also fed them to a fast ADC to produce an analog ramp signal. The ramps went to the X and Y deflection inputs of a large screen oscilloscope and the data input and output pins went to the Z axis modulation input. By alternating the data lines logic levels the screen would turn all dark or all light but any errors were immediately visible as spots or lines where the storage had failed. A dot indicated a single cell failure and a line or block indicated an addressing failure.

Brian.

 

[BradtheRad]

Is the DIMM installed in a computer? Some computers have a ram diagnostic built in. (Call it by interrupting startup, then look in the bios menus.) In addition there are diagnostic utilities which test ram by sending various patterns of data. If it finds a fault, you see a message telling which location it is. This may also tell which chip it is, although you'll need detailed knowledge to know which chip should be removed and replaced.

 

[c_mitra]

What hardware test should I use to find out a faulty memory chip in DIMM module?

Your query is not clear enough.

You want to test a RAM that is going to be used in a DIMM for memory errors?

OR

You have a DIMM that has been reported with a memory error and you want to repair that by replacing the faulty RAM?

Most PCs have memtest software and the software solution is very good. But that takes time. Also in case you are in a commercial environment, this is very impractical.

But I am simply curious whether you have done any cost analysis for such repairs.

 

[Jane1]

I can check the booting (process) via COM port but ONLY if all 4 RAM chips are working. Yes, those 4 RAM chips form a complete DIMM like memory module.
If any of those 4 RAM chips is faulty the device will not start booting and I can see nothing as output (on COM port)
So, I would need to test which RAM chip is faulty only by hardware means - no software solution because CPU does not boot properly.
Any idea?

 

[betwixt]

If there is more than one DIMM in the system, you could try swapping them. Possibly by rearranging the memory map it will allow you boot as far as running diagnostics on the other DIMM. Other than that, the only other method is a dedicated diagnostic tool for the DIMM alone and as I stated earlier, they are VERY expensive, think in the $US 100K+ region!

Even if you could isolate one IC on the DIMM, you still need to find a matching replacement and you need SMD soldering equipment to change it.

DRAMS are complicated VLSI devices, there is no easy way to test them I'm afraid.

Brian.

 

[BradtheRad]

Windows machines which I've used have an option for slow or fast startup. The slow option tests ram. The fast option does not.

My laptop had green goo on one ram chip. It interfered with contact. I would be watching a video and then suddenly the machine froze. This happened several times.

I performed diagnostics. It reported a ram error. It was not always the same location. When I looked at the chips and saw the green goo, I realized that must be the culprit. It crept in between the contacts. It caused a ram error, and not in the same location each time.

Because I selected fast boot instead of slow boot, there is a chance the error was present in a memory location at startup, and did not cause a problem, until that memory location was accessed. Then the system froze.

When doing a ram diagnostic, it helps if you remove all ram except the one you're testing. Of course this should be done when the machine is off.

 

[Jane1]

@betwixt:I do not think that RAM chip swapping will help. Those RAM chips has 8 bit data bus and together they form 32 bit data bus for CPU to read/write that 32 DWord (Double Word) format data. So, I think that CPU checks if that written 32 DWord is the same as it is read - to check that ALL those 4 modules work properly.

 

[betwixt]

It can help sometimes. Although the RAM data bus is 32 bits wide, access can be made in 8-bits. Hardware (called the cycle conversion logic) on a standard motherboard can read and write four 8-bit values individually to one physical 32-bit wide address. In fact if you look at Pentium+ data sheets you will see that instead of low address pins they have "byte select" pins instead.

When first switched on, most BIOS do a memory test in 8-bit chunks although having established a base working area in RAM is good, they then start using 16 or 32 bit writes and reads to find out how much RAM is present in the machine. Whether it tests every address or just one every 64K or so is the difference in fast or slow booting mentioned by Brad in post #7.

A cell failure at a low address may stop the machine booting but if swapping the DIMMs moves the fail to a higher address it may leave enough functional RAM for a diagnostic test to run.

You have nothing to lose by trying!

Brian.

 

[c_mitra]

Although the RAM data bus is 32 bits wide, access can be made in 8-bits. Hardware (called the cycle conversion logic) on a standard motherboard can read and write four 8-bit values individually to one physical 32-bit wide address. In fact if you look at Pentium+ data sheets you will see that instead of low address pins they have "byte select" pins instead.

I am not so sure - most modern PCs do not address the RAM by bytes- a group of 4 or 8 bytes are called at a time. This reduces the width of the address and the number of references. I have also heard that the RAM is addressed page-wise- i.e., 4096 bytes in each reference.

It is the memory controller that talks to the CPU directly. So bits and bytes are not really a problem.

It is entirely possible that I am wrong but I still want to know.