An Explanation of the Impact of Optical Restrictions on Semiconductor Yield Rates

Like it says – Ian Anderson

There has been discussion over the past year or so over dwindling yield rates for all semiconductor manufacturers. Just so we are all clear on what this means, a yield rate is the ratio of working chips to total chips.

If a manufacturer can produce a chip for $5 and half of them work, the cost of making a working chip is $10. This is a 50% yield rate. If a manufacturer wants to shrink their process from 300nm to 250nm, their yield rate will drop if they use the same technology. To reduce the price of making working processors, they will then need better technology.

The production of semiconductors is a closely guarded secret and I will not pretend to know more about it than most of you. It is essentially photography, or to be more precise, photo reduction. An image of the components to form the layer is photographically etched and/or deposited onto the silicon. The process is then repeated to build up the layers of the processor.

This is not a discussion of physics, but when you are dealing with optics, there are some laws you need to follow. There are limits to how fine a line a lens can image. This is called the diffraction limit, where the wave nature of light will not allow higher resolution.

The simple solution is to increase the size of the lens. Doubling the lens diameter will double the maximum resolution. The other element to this equation is the wavelength of light you are using. Once again, the equation is simple – a wave of half the length will double the resolution. The big problem with this is shorter wavelengths of light are more easily absorbed and scattered by glass, thus reducing yield rate.

The purity of the lens element (glass) also has an impact on yield. Every inclusion and defect in the lens will show up as a defect in the process. Purity is measured as the number of particles per unit volume. Increase the volume and you increase the number of inclusions. A semiconductor fabrication process has an absolute limit on the number of inclusions. As larger lenses are needed, the purity must necessarily increase.

The theory is simple. In practice it is next to impossible. To keep the numbers simple, I will assume both doubling the lens size and halving the wavelength of light. This necessitates the optical surface be twice as good, doubling the price.

To double the diameter from 6″ to 12″ means quadrupling the area and the amount of optical finishing to be done. The new optic will cost eight times as much as the old one and will perform four times as well. Doubling both factors squares the resolution improvement but cubes the cost.

I honestly don’t believe fabrication cost is the problem. The problem is the optics semiconductor manufacturers want cannot be made. It is impossible to increase purity to the levels necessary for semiconductor fabrication in any larger sizes. It is nearly impossible to make better optics than are currently available.

Companies now use marginal optics causing lower yield rates.

At this point, it may become cheaper to use more silicon than to shrink the transistors. The problem will be solved. But it will not happen as quickly as it has in the past.

Ian Anderson – Custom Optical Systems