The shortest distance and chip fabrication

Posted on September 29, 2005 , filed under Stocks | Print This Post

Take a piece of paper and draw two points on it, Point A and Point B. Now try to find the shortest distance between the two points. It’s a straight line right? Well not exactly. I’ll show you another way to find an even shorter distance between the two points. Fold the paper in between the two points. Now that the two points are almost on top of each other the shortest distance is still a line, but a line that moves along the Z or depth plane instead of just the X and Y planes of a flat surface.

Tradition chip fabrication has always been thought of on the X and Y planes of a flat surface. These days it’s getting harder and harder to minimize the paths a chip needs to communicate. New techniques, beyond new process techniques in minimization, will be required to increase the processing power of today’s CPU’s.

Above, you see a cross section of an AMD Athlon X2 CPU. What you can see in the picture are the two CPU cores sitting next to each other and the parts of each core. What you can’t see without proper magnification are all the tiny paths necessary to allow the cores and the parts of each core to communicate with each other. Since Moore’s Law was first introduced, the primary method of increasing the speeds of a CPU have revolved around shrinking the parts and distances between the parts to communicate. Imagine now, that instead of designing CPU’s on a flat surface, you placed the cores of a dual CPU or the parts of a single CPU on top of each other. Doing this, you would massively decrease the distance between the points on a chip that are required to communicate. This technique is known as stacked chip design. At the moment, there are only a handful of companies that have the required expertise and tools to create stacked chip designs.

Stacked chip designs require special software tools to design on the Z plane and tools to calculate heat dissipation. Heat will be a major problem to tackle with parts sitting on top of each other instead of adjacent to each other. If stacked chip design takes hold, an entire industry will form around the process. At the moment stacked chip design and multicore CPU’s are the leading candidates in advancing Moore’s Law. I predict that nanotechnology developments will ultimately advance chip design and performance, but not cost effectively for at least another decade. At the moment I see more novel and simplistic approaches, like stacked chip design, in speeding up chip communications. Early investors will look for opportunities with profitable companies that provide design tools, process technologies or equipment to the semiconductor industry.

I would keep my eyes on AMD, INTC, AMAT, SLR, SWKS, IBM, TXN, UMC and TSM as the major players that would take advantage of stacked chip design.

Add this post to your favorite social networking site