World's smallest SRAM memory cell announced by IBM

[Urian]

A critical component for high-speed computer memory that is about ten times smaller than those currently available was unveiled by IBM today.

The new IBM SRAM (static random access memory) cell is less than half the size of the smallest experimental cell reported to date, and ten times smaller than those available today. IBM says this new cell has the potential to give a major system performance boost for critical business applications like banking and digital media.

IBM says its researchers optimized the SRAM cell design and circuit layout to improve stability and developed several novel fabrication processes in order to make the new SRAM cell possible.

 

[PC-Engine]

How does it compare to Mosys 1T-SRAM-Q? Any actual cell size figures? Any timeframe when this will actually be achievable in volume manufacturing?

 

[randycat99]

It doesn't compare. The only thing "SRAM" about 1T SRAM is that they used it in the name.

 

[Fox5]

It doesn't compare. The only thing "SRAM" about 1T SRAM is that they used it in the name.

I thought they used SRAM in the name because it had certain properties normally attributed to SRAM.(low latency, low power consumption, and can hold a charge for a sustained period of time)

 

[randycat99]

That is certainly the desired mental image to be associated in naming the product as they did. Is it directly comparable and replaceable in performance with genuine SRAM? Not in the least. That did not stop various videogame review sites from eating it up for all hype's worth (think "Blast Processing"). It's simply an alternative to fast VRAM in an embedded format, rather than the 2nd coming for the stuff they really use in CPU L2 caches, for instance.

 

[Li Mu Bai]

That is certainly the desired mental image to be associated in naming the product as they did. Is it directly comparable and replaceable in performance with genuine SRAM? Not in the least. That did not stop various videogame review sites from eating it up for all hype's worth (think "Blast Processing"). It's simply an alternative to fast VRAM in an embedded format, rather than the 2nd coming for the stuff they really use in CPU L2 caches, for instance.

The sites I read said it was comparable, though not equal to SRAM in some of its functionality. It indeed still required refreshing, as it was at its core DRAM, the difference was how it went about acheiving it. 1T-SRAM's design can mask the refresh process quite effectively to the point where they can claim latency and bandwidth figures roughly equivalent to those of standard SRAM's. Do you have bencharks or comparative numbers to counter MoSys's claims?

 

[randycat99]

That's the thing- if you do a search, everybody is literally touting the same figures and headline (almost word for word) as the original MoSys announcement. Nobody has bothered to really find out or verify the claims. It is essentially a marketing blitz campaign turned "fact". If you read closely though, you will find the occasional mention that the design is essentially just a faster variant of regular ole DRAM packaged and processed to fit slightly different design criteria. They could have named it "really fast DRAM on a chip", but that wouldn't quite have the same marketing oomph as calling it a form of "SRAM". It's water under the bridge at this point, and doesn't make a GC any slower or less enjoyable as before. It's not even worthy of a new topic, let alone as a derailment to this one about this new IBM SRAM.

 

[Inane_Dork]

Estimated time to market?

 

[pcchen]

Since it uses "mixed e-beam and optical lithography" it probably won't be available any time soon.

 

[StefanS]

That's the thing- if you do a search, everybody is literally touting the same figures and headline (almost word for word) as the original MoSys announcement. Nobody has bothered to really find out or verify the claims. It is essentially a marketing blitz campaign turned "fact". If you read closely though, you will find the occasional mention that the design is essentially just a faster variant of regular ole DRAM packaged and processed to fit slightly different design criteria. They could have named it "really fast DRAM on a chip", but that wouldn't quite have the same marketing oomph as calling it a form of "SRAM".

To be fair MoSys never claimed that their 1T-SRAM would be a 6T-SRAM in the traditional sense, all the said was that it had similiar qualities while lower error margin and thus:

1T-SRAM provides significant advantages over traditional SRAM in density, power consumption and cost that enable designers to more economically use a larger amount of memory. Instead of six transistors utilized in a traditional SRAM storage cell, each 1T-SRAM storage cell contains only one transistor and one capacitor, thus reducing the silicon required and lowering cost.

http://www.mosys.com/products/1t_sram.html
Try the technical article pdfs for details on it; pretty nice & easy to read.

 

[randycat99]

Of course they did not make any explicit claims. That's the catch- by naming it "SRAM", a meaning is implied for those who would not think to investigate further. Now no one is going to question it, since we have 3 zillion listings in Google that are essentially variations of the initial announcement, would they?

 

[StefanS]

Of course they did not make any explicit claims. That's the catch- by naming it "SRAM", a meaning is implied for those who would not think to investigate further. Now no one is going to question it, since we have 3 zillion listings in Google that are essentially variations of the initial announcement, would they?

Well, those who want to talk about specifics of consoles should at least try some basic research; starting with the manufacturer of its parts. It's a real shame that all most all of GCN hardware review/faqs are poorly done and poorly researched

 

[randycat99]

Very true, my friend. I once had a discussion with somebody who thought they had an "infinite Q" treble booster in their fancy audio system. It's an everyday quest to get by the legions of spec chasers and buzzword fanatics which exist for just about anything that is a consumer electronics product.

 

[PC-Engine]

Heh something bothering you Randy? Randycat99, the guy that claims PS2 is capable of more GFLOPS than GCN....

PS2 = 6.2 GFLOPS + imaginary number > GCN = 10.5 GFLOPS

EE/GS = imaginary image enhancing algorithm = better image rendering for televisions

Where's those imaginary GS GFLOPS Randycat???

It doesn't compare. The only thing "SRAM" about 1T SRAM is that they used it in the name.

Show me a link where it proves 1T-SRAM-Q is nothing but marketing with facts and figures...or do you want us to believe it's all marketing just because YOU say so?

Oh and btw the term is constant Q equalizer.

http://www.rane.com/note101.html

Still waiting for the paint to dry Randy?...

 

[randycat99]

Yeah, and you thought it was behaving as an "infinite Q" device. Yes, constant Q exists, but you still to this day do not understand what it really does.

If you are content to believe that 1T SRAM is simply SRAM, go right ahead. It's water under the bridge- just doesn't build much on your credibility.

WRT to the PS2/GC GFLOPs discussion, get over it. To this day, you appear to be confused on what was really being said, apart from your deliberate misrepresentations of meanings and contexts. Just let it go- it only makes you look more foolish in clinging to that as some sort of "defense".

Your 1T SRAM question was simply ridiculous. Aside from that, I hope you will let this topic continue unmolested on its true topic- IBM's SRAM.

 

[PC-Engine]

Yeah, and you thought it was behaving as an "infinite Q" device. Yes, constant Q exists, but you still to this day do not understand what it really does.

Actually that's what basically happens when you increase the gain. Just look at the bell curve.

Funny how you thought every EQ was desgined the conventional way where it's not possible to boost a very narrow frequency ie 20kHz just like you thought the PS2 GS also had GFLOPS since the gpu in GCN had GFLOPS and just like you thought the EE/GS improved the image of a tv when in fact it's only being used as a XMB GUI generator, the same one used in PSX.

Must be getting extremely hot (steaming) in that coffin...

If you are content to believe that 1T SRAM is simply SRAM, go right ahead. It's water under the bridge- just doesn't build much on your credibility.

Oh man what a riot!!! You of all people talking about credibility?

How credible is (EE) 6.2 GFLOPS + (GS) imaginary GFLOPS > 10.5 GFLOPS???

Actually I asked how the Mosys 1T-SRAM compared and since you don't know yourself you assume it's just marketing with ZERO evidence to back it up. Not really surprising though, considering you got nice smack downs in the past and again yesterday then again right now.

WRT to the PS2/GC GFLOPs discussion, get over it. To this day, you appear to be confused on what was really being said, apart from your deliberate misrepresentations of meanings and contexts. Just let it go- it only makes you look more foolish in clinging to that as some sort of "defense".

Everytime I see one of your superfluous posts with no substance to back it up it reminds me of the good old days at GTForums where your equally cluelss friends came to try and back you up to only get beat down with real cold hard evidence. Nice to see you behaving at Beyond3D where people here understand and can appreciate PowerVR technology.

Your 1T SRAM question was simply ridiculous. Aside from that, I hope you will let this topic continue unmolested on its true topic- IBM's SRAM.

Wait...so you claim the question is rediculous to only backpedal and offer your WORD that it's purely marking? Riiiiiight. Who are you anyway...God???

 

[pcchen]

Since this SRAM cell is not going to be avaiable for some time, there is really no point to compare it with a technology already available on market. Anyway, 1T-SRAM-Q normally are quoted as 0.50 micron^2 or 0.28 micron^2 cell sizes w.r.t. 130nm and 90nm process. The IBM's cell is 0.143 micron^2, for 32nm process or beyond.

Of course, cell size is not the same as average bit size because of difference in overheads. For example, a 1T-SRAM-Q macro in 130nm may have an average bit size of about 1.2 micro^2. The overhead of SRAM macros will be somewhat smaller.

 

[PC-Engine]

Thanks pcchen, I knew it was too good to be production ready. Heh 32nm...

 

[Li Mu Bai]

Yes randycat you labelling it as "blast processing" makes it quite clear you believe it to be nothing more than a "marketing ploy" or myth. Yet you have no supplementary evidence that states the contrary. Other than your opinion, which frankly doesn't hold up as PC-Engine has so thoroughly pointed out. Funny thing is MoSyS never said that it was standardized SRAM, it just acheived SRAM's sustained latency numbers, etc. Nintendo nor MoSyS was touting this as some form of advanced tech. that made the GC a more powerful platform, as the average consumer doesn't know the difference regardless.

 

[StefanS]

Yes randycat you labelling it as "blast processing" makes it quite clear you believe it to be nothing more than a "marketing ploy" or myth.

Frankly, both you and PC-Engine are misunderstanding him. Fact is 1T-SRAM is not a SRAM by convention. Randycat never said that it is standard conventional DRAM either. He was mainly complaining about the uninformed handling of buzzwords and the way 1T-SRAM was introduced into a topic which it is not really related to. In my opinion he's right on both assertions.

As on the properties (read my link, above) 1T-SRAM enjoys similiar qualities. It has low latency, although it doesn't have SRAM latencies. Li Mu Bai. Its advantages over SRAM are mainly lower error rates, cheaper production, less heat etc. Anyone who wishes to know more about it should read MoSys's technical articles (believe me, they are very easy to understand).