Every two years, someone takes a look at how the semiconductor business is doing and estimates how long we can continue in the same fashion, making transistors smaller, budging them in closer together and making our CPUs work at lower and lower voltages. Physicist Michio Kaku now says that Moore’s law, which dictates the speed at which we can increase the amount of transistors in a given area, will come to and end in less than ten years.
Moore’s law is a theory by Intel co-founder Gordon Moore, which states that the number of transistors that can be crammed into a given space would double roughly every two years. Moore set down the law in a paper he wrote in 1965 concerning the acceleration of processing power. The increases come from the nanomater process (nm), which is how semiconductor manufacturers like Intel, Samsung, Micron and formerly AMD close the gap between the transistors in their designs to make the most efficient use of the space they’re given.
Moore’s law, however, will eventually run into a physical issue, where we can’t possibly make smaller transistors. “We’re already manufacturing on a microscopic scale”, says Michio Kaku. “At some point in time we’re either going to run into cost issues or we’re going to be limited by the laws of thermodynamics and quantum physics. We’re going to have to make them out of something else, like DNA or Carbon molecules!”
AMD’s Chief Product Architect, John Gustafson, appears to agree. Gustafson says, “We’ve been waiting for that transition from 28nm to 20nm to happen and it’s taking longer than Moore’s Law would have predicted…I’m saying you are seeing the beginning of the end of Moore’s law.”
However, that’s not the entire story. AMD has contracts with two fabrication companies that make their chips for them – Global Foundries (formerly a part of AMD) and Taiwan Semiconductor Manufacturing Company (TSMC). TSMC can manufacture AMD’s processors and GPUs at the 28nm process and in fact they also do fabrication for Nvidia as well. Global Foundries, until recently, couldn’t progress beyond 32nm because of cost issues.
This is a stark contrast to how other companies are currently doing. Intel has taped out their process of producing chips at 14nm and will be using it to make the Haswell successor, Broadwell. Already their products are made with the 22nm process. Samsung is now at 19nm, Micron is currently at 22nm and Toshiba trails at 19nm as well.
The companies that are making enough money on their products are the ones ahead in the game to make them smaller and more efficient. However, Intel is the only one with enough money to keep Moore’s law progressing for their products. With time, Samsung will be a close second.
The main advantages of using a smaller process, even if you don’t change anything about your chip design itself, is reduced power consumption and electricity leakage. The benefits mean the chip can be scaled to fit better into thermally-limited form factors like laptops and tablets, which typically rely on cooling that’s just enough to keep things in check. However, there is increased heat generation because everything is spaced so close together, so it’s a very fine line to tread.
In addition, moving to smaller process nodes (same meaning as fabrication process) is also extremely expensive. Not only do you need machines and manufacturing processes that are more accurate and efficient, your methods may even take anything from a few months to a year to perfect to get good die yields out of the process – a stumbling block that Nvidia had with TSMC when their Kepler architecture wasn’t being made error-free with a low enough cost-to-benefit ratio to make it worth the risk.
The Geforce 600 series was delayed for over five months because of manufacturing issues.
AMD may be right about Moore’s law coming to its end as we know it, but it’s more likely that they’re using this to divert attention away from the fact that they have poor relationships with other, cheaper and better fabrication plants.