Published in News

ARM And TSMC Sign A Deal To Manufacture 20nm 64-Bit ARM SoCs; Still Behind Intel By 3-Years

by on24 July 2012 2192 times
armchip displayv2

At their earnings call the other Day Intel made the statement that they are at least three years ahead of other companies when it comes to silicon manufacturing. Yesterday and today the news is all about how ARM is preparing to compete with Intel in the server market with their 64-bit RISC processors which are to be manufactured using the same tri-gate manufacturing that Intel is currently using for their Ivy Bridge CPUs. You know the ones that have been shipping since early this year? This means that no matter how thin you slice it ARM will not have able to make a tri-gate CPU until at least 2014-2015.

So what is all the fuss about? Most of you will know who ARM is especially if you own an iPad, iPhone, Smartphone, Tablet etc. as ARM IP is inside them. All of these will be under different names (Tegra, Snapdragon OMAP, A5), but they all use ARM IP. An ARM CPU is, in simple terms, a small, low-powered, RISC (Reduced Instruction Set Computing) CPU. This means that it uses a very simplified and optimized set of instructions (often as many as a Complete Instruction Set CPU) which reduce the amount of work any instruction uses at one time. It makes it very easy to optimize your code to work efficiently and quickly with your CPU.  RISC processors are also tend to have less transistors in use for core logic and use less power than their CISC counter parts.  One of the most famous RISC CPUs is the DEC Alpha. This RISC CPU ended up being the genesis for the AMD Athlon CPU which changed the computing world in more way than one.

Because smartphones and tablets have requirements for smaller form factors and lower power usage ARM’s choice in using RISC was a smart one. We are also guessing they banked on the fact that Linux, Apple, Microsoft and a few other companies already had experience working with RISC implementations for their operating systems (Apple in particular took advantage of this with iOS). However, all is not pizza and beer with RISC. Because of the reduction in the way that instructions can use CPU time drivers can be problematic. Anyone that used Either a DEC Alpha system or a Mac back in the PPC days (From Motorola or IBM) can attest to some of the complications even adding more memory could have, the idea of simple plug and play drivers was something that took a while to get nailed down and unfortunately, by the time Apple had it working well they moved back to the CISC world of Intel’s x86 CPUs.

Now what happens when you try to push this concept into the non-mobile world? Well there are a couple of obstacles that will hurt you. The first is memory architecture which ARM is addressing (pardon the pun) with their 64-bit instructions that they released last year. Next is to work on the manufacturing process to increase transistor density and lower power usage and leakage. To do this ARM has signed a deal with TSMC (Taiwanese Semiconductor Manufacturing Company) for an exchange of technology. This exchange should allow ARM to design their new processors faster and to fit the manufacturing process it will also allow TSMC to improve their yields and fine tune their manufacturing to suit ARM designs.

TSMC is working on validating their 20nm process right now, but are not near production. We would not expect 20nm to start rolling out before mid-2013 with no word on the next die shrink at all. Intel is already focused on 14nm and is shipping 22nm in volume. Even with best case scenario do not expect 20nm 64-bit ARM SoCs before late 2014 early 2015. By that time Intel may have already dropped their 6 watt Centerton Sever chips on the market. This would still leave ARM behind unless they can find a way to trump the performance lead that Intel has with their x86 products. We also have to consider the implications of ARM based servers and the uses they would be put to. What items in the data center would benefit from this? The argument has been made for web servers, but in the end most companies will virtualize those using VMWare, Citrix or another hypervisor style virtualization platform. This way they are still only paying a fraction of the overall server costs per “server” as you can run a significant number of guests inside even a moderately equipped x86 server from AMD or Intel. As always your main power draw in the data center will come from the massive storage arrays that are needed these days to keep all of the data that your average enterprise needs.

Getting into the software side we there are even more obstacles as companies will need to change their software to work with the RISC based ARM CPUs. This can be an expensive proposition if you do not already have the tools to do it. As we mentioned there are companies that do, but we are not sure if they are going to want to extend this into the server market where the ROI (Return on Investment) will be very light. No to put it simply ARM has a long way to go before they get into the datacenter in any real or meaningful way and by that time we would expect by AMD and Intel to have products to match or exceed the performance that ARM can offer.

Discuss this in our Forum

Last modified on 24 July 2012
Rate this item
(0 votes)

Leave a comment

Make sure you enter all the required information, indicated by an asterisk (*). HTML code is not allowed.