Dom

MAJOR FOUNDRY TSMC said today that it's going to use PDF Solutions' Yield Learning Infrastructure, underlying the difficulties semiconductor firms face as they move to the tinier process technology.
That's something that TSMC acknowledges too. In a statement, it said that 90 nanometer technology involves design and process technology and creates "yield issues".

Basically, yield in the semiconductor world refers to how many reliable working dies can be made from one silicon wafer. The better the yield, the more semiconductors can be sold, and the quicker the chip firm can start paying off the cost of the very expensive equipment required to make the shift.

Essentially, moving from .13µ (microns) to .09µ microns, or 90 nanometers, requires a level of chip engineering that's far greater from the earlier shift from .18µ. When TSMC attempted that move, it discovered that it wasn't able to easily shift to the smaller process then, causing difficulties for a number of customers, notably Transmeta, Via and Nvidia.

TSMC estimates that its work with PDF means that the joint project will be completed in the fourth quarter of this year, at its 12-inch (300mm) Fab 12.

A 12-inch wafer obviously allows more chips to be "printed" but the kit to move away from the widely used eight inch silicon wafer manufacturing process is also highly expensive.

Intel is also cautiously beginning its move from the current .13µ techniques it uses at its fabs to 90 nanometers, but the shift will probably be more gradual than the earlier shrink.

It, and other semiconductor firms, hope to successfully move to larger wafer sizes and to smaller die sizes, thus maximising the number of chips it can sell.

But now the technology is shifting to the nano level, it means the design and manufacturing challenges are that much greater, as the current technology pushes at the limit of Moore's Law.

The current state of the IT industry also poses more serious problems. As prices fall, and PC sales still continue to lag, that means there's less money to invest in capital equipment. With a fabrication plant (fab) costing a cool $2 billion or so, revenues and profits become ever more important, and there is serious concern that the money just isn't going to be there unless people believe there's a need to upgrade their machines.

While there will always be a certain proportion of buyers who want the latest and the greatest, many aren't convinced that an extra bit of megahertz here or there will really make their "computing experience" that much better. Right now, the industry is pinning its hopes on Microsoft's Longhorn and PCI Express to give people - and we mean the mainstream, not the enthusiast - a really good reason to buy.

_______________________
Source: TheInquirer