Toshiba offers 130-nm process as foundry service

KnightBreed

Newcomer
From the article at EEtimes:
Toshiba's 130-nm technology employs eight metal layers and has a standard gate length of 110 nm. DRAM macros for 4, 8, 16 and 32 Mbits are available, and multiple DRAMs can be embedded on one chip. A 6T SRAM cell measuring 2.48 square microns is also available. Copper wiring combined with dual-damascene technology makes for a minimum wiring pitch of 0.34 micron, the company said.
Infineon dumped its .17um embedded dram service, and Toshiba now offers their .13um lines to third parties. Nappe1 mentioned in the huge thread in the 3D hardware forum that Bitboys have already partnered up with a new foundry for future development. Toshiba, perhaps?
 
hmmmm interesting
but wonder if the bitboys chip is still going to be as competitive now that R300 is here and NV30 is around the corner
 
To step away from the Bitboys discussion:

With another company competing with the industry's biggies UMC, TSMC, IBM, Chartered, etc, how will Toshiba effect the industry as a whole? Are mask and wafers prices affected by competition?
 
Nappe1 said:
I can surely say that Bitboys new fab partner is not Toshiba. ;)

Thanks for clearing that up Nappe1 ;)

I knew you couldn't say who it was, but I at least got you to crush all doubts of who it wasn't 8)
 
> With another company competing with the industry's biggies
> UMC, TSMC, IBM, Chartered, etc, how will Toshiba effect the
> industry as a whole? Are mask and wafers prices affected by
> competition?

For Toshiba, IBM, TI, Fujitsu, NEC, and LSI (and several others), the wafer-fab business is a business segment within the entire company.

TSMC and UMC are 'pure-play foundries', which means their entire (or majority) revenue comes from fabricating wafers for others. (Actually UMC owns a few chip companies, like Mediatek.) From what I have *heard*, TSMC and UMC offer the best piece/price (price per wafer.) Both foundry run very MINIMAL in-house design services (place&route, built-in-self-test circuitry), choosing to focus only on wafer manufacturing. TSMC and UMC are best suited for companies who want to produce an ASIC in large volumes (millions.) Both TSMC and UMC offer very similar product (in terms of process-line.)

The other companies I mentioned all run 'design services' for its customers. This can include back-end work (standard cell placement and interconnect routing), drop-in IP (intellectual property, like a DDR I/O cell or gigabit-serial block), full testing-automation (includes the built-in-test-circuitry and wafer probe service), and/or packaging (this is tied with the back-end work, because of the I/O cell placement.) To subsidize these service businesses, the customer's per/wafer-pricing is corresponding higher (compared to TSMC/UMC, IBM and TI charge 2.0X or more for a similar digital-logic process.)

Some foundries also offer 'exotic' (and expensive) manufacturing processes, like GaAs or SiGe (actually, TSMC and UMC also offer SiGe.) These are crucial for high-speed RF (radio frequency) circuits and specialized analog circuits, like 10gigabit ethernet, etc, but for very large digital ASICs, they aren't price-competitive with standard silicon due to higher defect rates.

Given a choice, and all other things being equal, I think customer will pick the choice with lowest cost. Aside from cost, there are other factories. For example, Qlogic and Emulex are both LSI customers, probably because of LSI's gigabit transceiver core. The availability of LSI's gigacore enabled both Qlogic and Emulex to simply drop in a transceiver macro (and not worry about its design/verification.) Currently, Qlogic is actively looking to switch away from LSI, due to LSI's entry into the SCSI market.

...

that said, the merchant foundry industry is currently in a state of 'over-supply' (too much fab capacity), obviously caused by the high-tech decline in the communications industry.

This is the opposite of year 2000, when both TSMC and UMC were turning out wafers at 100% capacity. Both TSMC and UMC had to turn away a lot of 'smaller volume' customers/

** EDIT **
A correction to a potentailly inaccurate/misleading statement -- "compared to TSMC/UMC, IBM and TI charge 2.0X or more for a similar digital-logic process."

IBM and TSMC have fundamentally different pricing-structures. TSMC delivers fabbed wafers to the customer. The customer pays per wafer. IBM (and many other merchant foundries) deliver packaged and tested chips (i.e. the wafer is cut, the individual dies are tested, and only the good ones are saved for packaging and final delivery.)

From a business standpoint, TSMC's customers share a great deal of manufacturing risk. Whether a production run has high yield or low yield, the customer pays the same fixed rate. IBM shoulders most of the risk by delivering only known good dies.

With either business model, profit maximization occurs with high manufacturing volumes of a single design. In other words, "1 customer who orders 9999 wafers" is better than "9999 customers who each order 1 wafer." So don't think for a second that TSMC doesn't care about its customer's wafer yields -- TSMC profit is just as dependent on prolonged (and successful) production runs as IBM's.
 
PC-Engine said:
Nappe1 said:
I can surely say that Bitboys new fab partner is not Toshiba. ;)

Thanks for clearing that up Nappe1 ;)

I knew you couldn't say who it was, but I at least got you to crush all doubts of who it wasn't 8)

...but I didn't say how close the right one it was... ;)
they now have fab partner who has earlier experience from normal GFX asic as also using embedded memory technologies for GFX chips. their new fab partner also seems to push eDRAM development very aggressively forward.
 
they now have fab partner who has earlier experience from normal GFX asic as also using embedded memory technologies for GFX chips. their new fab partner also seems to push eDRAM development very aggressively forward.



hmmm i seem to have an very good idea stuck in my mind....can't seem to get it out
maybe later
 
well, I am not sure if they announce it when their new website goes online, which should happen in a few weeks.

but that latest tip is the closest thing that I can comment. There shouldn't be many choices left.
 
ahhhhhh i just can't seem to get it out of my head

ok i love these guessing games
so they have made gfx cards before with eDRAM experience ...their website goes on in a couple of weeks so they don't exactly have a site at the moment....
well ca't be UMC they have a site and a section dedicated to talking about eDRAm but they havn't made any graphics card before.....nor have TSMC

damn only one i can think of is Mircon and ......
oh yes...how about NEC????????
is that it is that it????
 
for the answer, you have to just wait for official announcement. ;)

for your joy, you can continue guessing game, but be sure that you read right my hints. I was extra carefully while choosing the words.

Pascal and Phynicle: no comment. ;)
 
ok so Nappe has no comment to both our guess pascal
so who do you think it is pascal
NEC or TSMC given that TSMC has no previous experience in their own gfx
 
http://www.eetimes.com/story/OEG20011205S0054
i'm betting on it's NEC's full metal EDRAM process

The full-metal DRAM technology has the added advantage of requiring only a 500°C thermal budget, Kishi said. The process was created by adding between six to nine additional masks for a 0.13-micron pure logic process, which features copper interconnects and a low-k dielectric.
 
Maybe NEC. Maybe the japanese want to takeover the world :)

I think I will read all Nappe1 posts to try a better guess.

In the end maybe we will have to put him in a corner and BOOM :LOL:
 
Back
Top