In the datasheet of the Samsung KM4164B it says that after powerup it needs a 100us pause followed by 8 RAS cycles before proper device operation is achieved.
Not sure if this is related to the above but I see erratic behaviour when I use these as a 4116 (lower RAM) substitute.
They seem fine when I use them as upper 32K.
So I suspect that (ignoring the speed or timing) there are differences between the many brands and/or versions of 4164.
I just checked the Mostek 4116 datasheet and it also needs 8 refresh cycles.
In the datasheet of the Samsung KM4164B it says that after powerup it needs a 100us pause followed by 8 RAS cycles before proper device operation is achieved.
Doesn't sound unusual... The chip will probably need a little 'housekeeping' time after its power supply has stabilized. And a DRAM chip likely has some sort of state machine in it, which (much like a CPU) needs some cycles to initialize properly.
1st condition is met by the machine's reset signal. Since /RAS cycles result from Z80 memory requests (any memory!) and refresh cycles (and from the ULA as well in the case of lower 16K), 2nd condition is met after the Z80 runs through a few ROM instructions. Read: 'long' before it does a RAM test.
The usual argument goes: The problem with 4116's is that when they go pop, they often take the power circuit with them. Which can take out the 4116's after you've repaired them :-|
Edit: well I guess if you've replaced TR4 with a robust alternative you may be safe.
Replacing them when they die with something more stable that also doesn't require multiple power supplies is perhaps a good idea. Well it's a choice :-)
Not really... It's a pretty safe guess that many working 64K*1 chips can still be found when virtually all 16K*1 chips have died. For one, because of the 4116's multiple power supply issues. Add a few years of age difference, and the fact that many more of those 64K*1 chips were produced.
So if there's a choice, it's between "let your Speccy die" or "find alternative for the RAM". Not that RAM or ROM will ever be a big problem imho - ULA's might be though.
So if there's a choice, it's between "let your Speccy die" or "find alternative for the RAM". Not that RAM or ROM will ever be a big problem imho - ULA's might be though.
An identical problem was put up once by Sinclair: let your Speccy die or modify the voltage convertor to meet issue 4 standards. The official solution then was to wait until a Spectrum went in for repair. :-)
Comments
Guessing by the product id's you may have 300ns and 150ns devices? (On the train so can't check ! )
Not sure if this is related to the above but I see erratic behaviour when I use these as a 4116 (lower RAM) substitute.
They seem fine when I use them as upper 32K.
So I suspect that (ignoring the speed or timing) there are differences between the many brands and/or versions of 4164.
I just checked the Mostek 4116 datasheet and it also needs 8 refresh cycles.
1st condition is met by the machine's reset signal. Since /RAS cycles result from Z80 memory requests (any memory!) and refresh cycles (and from the ULA as well in the case of lower 16K), 2nd condition is met after the Z80 runs through a few ROM instructions. Read: 'long' before it does a RAM test.
I can't get many NEC D4164's so need to get other brands as well.
However I think that 4116's should be replaced by 4116's ;)
I have them in stock too.
Edit: well I guess if you've replaced TR4 with a robust alternative you may be safe.
Replacing them when they die with something more stable that also doesn't require multiple power supplies is perhaps a good idea. Well it's a choice :-)
So if there's a choice, it's between "let your Speccy die" or "find alternative for the RAM". Not that RAM or ROM will ever be a big problem imho - ULA's might be though.
I tested them and they work fine as upper (32K) RAM replacements in my Spectrums.
I'll test them as 4116 lower (16K) RAM replacements on my memory adapter pcbs this week and let you know the result.