Memory module mod anyone?

[timbob2469]

There is an article over at Anandtech.com called ,"OCZ 3700 GOLD: Building High-Performance Memory". As the title suggests, OCZ is making their memory faster. Making as in modifying to some extent. According to the article found here They are "MAKING" the OCZ 3700 GOLD out of screened, basic, Samsung (DDR400) chip modules. Seems they have a process that they are not even hiding (see the article), they basically cut 7mil off of the face of the chips for better heat dissipation. According to the charts found in many locations, 1mil = .001" . So .007" has to be removed from each side of the module (chipface-where the markings are). I am willing to sacrifice 1 stick of DDR memory to prove this out. I'm going to look through the Sunday newspaper supplements (circulars or flyers depending what country you live in) and get 1 stick of the cheapest DDR I can buy. I'll take it to work Monday and carefully beltsand to the required thickness. Of course I'll benchmark it Sunday for highest FSB and at different timings and measure the exact thickness using my trusty Mitutoya 0"-1" digital micrometers (.00005" increments). I'll also mask off the gold plated contacts before sanding on a 8" wet beltsander using 1000 grit belt. ( I machine glass for a living-cutting, grinding, drilling, lapping, polishing). This mod shouldn't be that hard to do, it just requires the right equipment. It reminds me of doing a port and polishing job on an intake manifold and cylinder heads in some way (see-building a hot rod). This is the stuff guys, do like the big guns in your own garage. I'll post back no matter what the outcome. Hell , I have a stick of Sammy in my backup rig that is getting pulled and benched on the RDA tonight, the backup will just have to run on 256MB .
Bob

[Áedán]

If you want to aid heat dissipation of a standard chip package, the best thing you can do is attach a heatsink! With no heatsink on the chip, you can expect a thermal resistance (junction to air) of around 50-60C/W. Putting even the most basic heatsink on the chip will drop that down to about 5-10C/W. (Basic heatsink in this case could just be a U shaped bit of metal)

Ok, I'm very skeptical about the claims.

[Kaitain]

Originally Posted by Áedán

If you want to aid heat dissipation of a standard chip package, the best thing you can do is attach a heatsink! With no heatsink on the chip, you can expect a thermal resistance (junction to air) of around 50-60C/W. Putting even the most basic heatsink on the chip will drop that down to about 5-10C/W. (Basic heatsink in this case could just be a U shaped bit of metal)

Ok, I'm very skeptical about the claims.

It's curious, but it would work to a very feeble and limited extent. Electrical grade silica substrate has a thermal conductivity of about 1.46W/(m^2.K) which is not a lot. I did a quick bit of maths with the following assumptions:

* Normal substrate thickness is 0.015"
* Final substrate thickness is 0.015" - 0.007" = 0.008"
* Surface area of a chip is about 1 sq cm
* Temperature inside the chip is about 35C
* Air temperature is a standard 25C (as you can tell, I prefer multiplying tens )
* All units converted to SI with appropriate errors

Using Fourier's law for conduction through solids, q' = kA(dT)/dx where k is the conductivity, A the area, dT the temperature difference and dx the limit of the thickness of the substrate gives these results:

For x1 = 0.015", the heat flux achieveable with a 10C temperature difference (more is better) is 0.09W
For x2 = 0.008", this obviously is almost double at 0.18W

Neither of these values are exactly impressive, so in any case you'll want to do as Áedán suggested and add heatsinks. I would imagine that, if you've already added a freon-cooled cryogenic ram cooler with integral beer chiller and A/C for a small room, then it might help you get another MHz. Personally I think it's a perfect way of throwing your money away.

It'll be interesting to see what timbob's results are...

[Áedán]

Originally Posted by Kaitain

Electrical grade silica substrate has a thermal conductivity of about 1.46W/(m^2.K) which is not a lot.

I believe we're talking encapsulation resin rather than silica substrate however.That might make some difference to your calculations, but I suspect that the thickness we're talking here isn't going to make a substantial difference.

[timbob2469]

Well , I'm holding off for a little bit before trying this ($75.00 is still $75.00- that's what I paid for the samsung pc3200). Maybe OCZ is accomplishing something else besides thinning the outer layer of the chips by their use of CO 2 laser ? Have you read the article ? On second look it just seems too simple and the information (or dis-information) about the process is just too easy to get.
Copied from the article," Samsung semi will provide die location information , we will be using the DP-7 CO 2 laser to remove the surface material from the package face only average laser cut depth is 7mill . After surface removal the package should be sent to ISE for environmental test and then re-speed graded after speed grade and classification chips will be marked ( standard process) and standard process thereafter."
I also run 2 X 256MB of OCZ pc3500 EL , on my main rig, they have heatspreaders and I'm not about to pull them off and void the warranty. But I did pull 1 256MB module out of my backup pc late last night, There are on-die components (capacitors, resistors? ) that are not that far below the surface of the chipface, so using a laser would allow thinning at the exact location of the chipface, my proposed method (beltsander) is probably too crude, I might try lapping them 1 chip at a time on a narrow strip of abrasive. Also I noticed these are the "D" chips, not the "E" type that they are using to make PC3700 GOLD from. So I think I'll get a stick of lower cost DDR memory.

[Áedán]

Most of the SMT components will be capacitors for decoupling purposes. There will also be an I2C EEPROM chip to hold the SPD parameters.

I'm pretty sure that the top layer of the chips is removed before being placed on any PCB. After all, it's cheaper to test the chips before they're placed on a PCB! That way you avoid having someone having to rework a PCB to replace a damaged/dead chip.

If you do decide to remove some of the resin, remember that the die internally probably faces upwards, and hence going too far will damage it.

[The Spyder]

reading your first post.... beltsander.... wet....

Not on ram you want to use after your done....

Dosent matter what grit you use, the force of the belt will rip chips off, get out of control, ive tried it on some old ram-
water=no

[Kaitain]

Originally Posted by Áedán

I believe we're talking encapsulation resin rather than silica substrate however.That might make some difference to your calculations, but I suspect that the thickness we're talking here isn't going to make a substantial difference.

Well if you happen to drop whatever the real thermal conductivity is into variable "k" above, the numbers will still work out in the same ratio. It's unlikely to be much different - typically solids that insulate electricity well also insulate heat well (there are exceptions). Either way you're talking about such low heat transfer capability to the outside world that a few microns more or less won't make enough difference to worry about.

Sounds like an expensive way of purchasing ordinary memory...

[Staz]

Originally Posted by Kaitain

I would imagine that, if you've already added a freon-cooled cryogenic ram cooler with integral beer chiller and A/C for a small room, then it might help you get another MHz.

Do you happen to have a guide to bbbbbuild one of those things? I'm not worried about chilling my RAM but the intergrated beer cooler would be a niceity.

[Kaitain]

Originally Posted by Staz

Do you happen to have a guide to bbbbbuild one of those things? I'm not worried about chilling my RAM but the intergrated beer cooler would be a niceity.

Funny you should mention that... I do have a concept diagram knocking around somewhere for a watercooling loop that also kept your beer frosty

[Áedán]

Well, a watercooling system that can keep your beer cool should be able to keep your coffee hot too.

I need to sort out a heatexchanger so I can put two loops in my system. Grabbing a mug and discovering that your tea has gone cold is not a pleasent experience.

[Kaitain]

Originally Posted by Áedán

Well, a watercooling system that can keep your beer cool should be able to keep your coffee hot too.

I need to sort out a heatexchanger so I can put two loops in my system. Grabbing a mug and discovering that your tea has gone cold is not a pleasent experience.

Waste heat recovery in a PC? This has to be done

[cloasters]

Originally Posted by timbob2469

Well , I'm holding off for a little bit before trying this ($75.00 is still $75.00- that's what I paid for the samsung pc3200). Maybe OCZ is accomplishing something else besides thinning the outer layer of the chips by their use of CO 2 laser ? Have you read the article ? On second look it just seems too simple and the information (or dis-information) about the process is just too easy to get.
Copied from the article," Samsung semi will provide die location information , we will be using the DP-7 CO 2 laser to remove the surface material from the package face only average laser cut depth is 7mill . After surface removal the package should be sent to ISE for environmental test and then re-speed graded after speed grade and classification chips will be marked ( standard process) and standard process thereafter."
I also run 2 X 256MB of OCZ pc3500 EL , on my main rig, they have heatspreaders and I'm not about to pull them off and void the warranty. But I did pull 1 256MB module out of my backup pc late last night, There are on-die components (capacitors, resistors? ) that are not that far below the surface of the chipface, so using a laser would allow thinning at the exact location of the chipface, my proposed method (beltsander) is probably too crude, I might try lapping them 1 chip at a time on a narrow strip of abrasive. Also I noticed these are the "D" chips, not the "E" type that they are using to make PC3700 GOLD from. So I think I'll get a stick of lower cost DDR memory.

It's probably the "screened" chips that make most of the difference. The cooling story just might be padding to help justify the high cost of this hand picked RAM! The idea of removing a few microns from the surface of the memory chips without an expensive machine shop at your disposal is humorous. Using a belt sander would be akin to adjusting the innards of a Rolex with a hammer.