The Birth & Evolution of Cortex-A9, and What’s Coming Next…

Just over 6 years ago, in October 2007, ARM introduced a product that had the potential to change the world. Not just in one market (like mobile), but nearly all known semiconductor markets – Looking back, we can now safely say this product did change the world for the better – the name of the product was ‘Cortex-A9’…

Press release from 3^rd October 2007 introducing Cortex-A9:

Technically speaking, ARM did announce 2 products at the same time: the Cortex-A9, a new single core processor, and the Cortex-A9 MPCore, a multi-core processor scalable up to 4 cores. The CPU itself featured a brand new partially out-of-order pipeline delivering 25% IPC improvement over Cortex-A8 in the same amount of power. The multi-core version additionally featured an integrated hardware cache coherency manager delivering up to 8000 aggregate DMIPS. Remember, this was at a time when processors like ARM11 with ~400MHz peak frequency (500 DMIPS) shipped in volume, so performance wise Cortex-A9 was superior to everything ARM had ever designed before.

It was clear fairly quickly that multi-core designs would rule the world, as most partners chose initially to implement (frequency balanced) dual core solutions based on Cortex-A9 MPCore technology instead of (high frequency) single processor solutions. This stopped a frequency race like we had in the PC world, and with scalability now a reality, designers started scaling across to quad core topologies. It did not take long until Cortex-A9 MPCore quad core became the highest performance solution available and is used in many (flagship) devices.

Initial Release & Revisions

The initial release in 2008 transformed rapidly into volume shipments and Cortex-A9 MPCore based silicon is available since 2010 across close to all market segments – Mobile (smartphones & tablets), embedded solutions, networking, automotive applications, any kind of consumer markets and even entry level servers, just to name a few.

Consumers have ever-increasing demands for performance at the same power or lower, so ARM and partners are innovating rapidly in response to those.  ARM is not standing still either, and the Cortex-A9 is a good example of our pace of continued innovation. Since the initial introduction of Cortex-A9 we have improved its performance by more than 40% (measured cycle-by-cycle). When rolling in the process node generations you can see that Cortex-A9 r4 is delivering a level of performance more than triple that of initial 40nm based designs with Cortex-A9 r1:

Even today we have partners taking full advantage of the latest improvements of the Cortex-A9 and are planning future projects scaling down to 20nm node and beyond.

Cortex-A9 translated into an immense success story for ARM and our partners – ARM expects to have shipped 1B partner units based on Cortex-A9 by end 2013.

What’s coming next…

There is only so much you can squeeze out of an existing processor design – improving on all edges until the product is close to the most that can be achieved with it’s base micro-architecture. With Cortex-A9 we have stretched those optimization over 4 revisions, and we saw a great opportunity to grab a clean sheet and start designing the next-generation processor which is going to take on where Cortex-A9 left off – so let me now introduce you to Cortex-A12.

Cortex-A12 is in a lot of ways very similar to Cortex-A9 – it is going to redefine new performance levels at the same level of efficiency that has made Cortex-A9 so successful, but this time focusing more specifically on the fast growing mid-range tier mobile and consumer market. The Cortex-A12 processor’s sophisticated ground-up pipeline brings a lot of features previously reserved to the high-end only now into the mid-range: Full out-of-order processing with a 11+ stage pipeline increases IPC performance by 40%+ over Cortex-A9 r4; tightly integrated VFP/NEON boosts performance by 50%; state-of-the-art prefetching algorithms and branch prediction technology help with complex workload demands of future devices. All of these improvements are coupled with power optimizations that enable the processor to match the power efficiency designers became accustomed to with Cortex-A9. Have a look here in case you are interested about finding out more details about Cortex-A12.

While we have Cortex-A12 on the way, with expected engineering samples from partners by 1Q 2014, we surely will see more Cortex-A9 based silicon coming, and who knows, maybe Cortex-A9 will reinvent itself all over again.