Apple tests Samsung 32nm process on A5-equipped iPad 2 and Apple TV

Apple gets a perfect test bed for the new process in low volume products.

Chris Foresman – May 4, 2012 1:02 pm | 54

45nm process A5 die (left) compared to 32nm process A5 die (right). Credit: Photograph by Chipworks

As noted recently by Chipworks, Apple is now shipping some iPad 2 models (and all third-generation Apple TV devices) with a 32nm A5 processor built on Samsung’s power-efficient “high-dielectric metal gate” (HK+MG) process. These products give Apple a relatively low-volume test bed to ensure its architecture works well with the process, and extensive testing by AnandTech shows that the 32nm process offers significant power savings over the older 45nm process used for processors in other iOS devices.

Apple’s A4, A5, and A5X processors, used in the most recent iPhone, iPad, and iPod touch models, are built using the 45nm process. Effectively, 45nm is the absolute smallest size that can be used to create a transistor or other circuit element on a chip using a 45nm process. This also affects the minimum distance between elements as well.

Moving to a smaller process, in this case 32nm, typically reduces the size of the circuit elements. This has two benefits: less power is needed to drive the transistors and other elements, and a particular chip design will take up less space. So a 32nm A5 is about 40 percent smaller than the 45nm A5.

It also uses less power to operate, but shrinking transistors beyond a certain size increases what’s known as leakage current. What happens is that the microscopic bits of a transistor get so close together that some current can “leak” through it even when the transistor is switched to “off.”

Comparison of silicon-on-oxide and high-dielectric metal gate transistors. Credit: AnandTech

To counteract that leakage current, Samsung uses a combination of a high-dielectric material and metal to create a more efficient gate. It’s more complicated to manufacture compared to previous gate compositions, which is why Samsung hasn’t used it until now. (By comparison, AnandTech said that Intel transitioned to HK-MG nearly five years ago.)

Moving to new chip processes often involves a lot of work, as problems are discovered during production and engineers have to find workarounds. This is the reason that Intel has used its “tick tock” strategy for the last several years—on the “tick,” Intel moves a current chip design to a smaller process. Then on the “tock” it introduces a new architecture at the same process. It appears Apple is employing a similar strategy, moving the proven A5 design to the new 32nm HK+MG process.

The benefits are actually three-fold for Apple. The Apple TV is a relatively low volume product, and the iPad 2 is being sold with both its original 45nm A5 and the updated 32nm A5. With this reduced volume, Apple can be sure it’s getting enough good processors without severely hampering overall production of iOS devices.

It also proves that the new process offers power saving benefits. Extensive battery testing by AnandTech reveals as much as a 20-30 percent increase in battery life for an iPad 2 with a 32nm A5 compared to one equipped with a 45nm A5. This offers Apple several options for the next iPhone: stick with the A5 and increase either battery life or performance, or build a 32nm A5X (or quad-core A6) that increases performance without a hit on battery life.

Finally, it saves Apple significantly on production. With the decrease in size, Apple can get a much higher yield of usable processors on each wafer. Each silicon wafer used to manufacture chips typically has a handful of flaws, and any chips that end up with those flaws get tossed. Apple is using another technique to increase yields as well—if the flaw happens to be in just one core, it disables that core and uses the chip in an Apple TV.