Google FeedBurner is phasing out its RSS-to-email subscription service. While we are currently working on the implementation of a new system, you may experience an interruption in your email subscription service.
Please stay tuned for further communications.
Get email delivery of the Cadence blog featured here
Which is more critical for you: simulation speed or accuracy? Can you have it all?
As consumers continue to thirst for more and better features in their electronic products, design complexity grows, silicon manufacturing costs increase, and competition rages on. Virtual prototyping offers one way to get your designs right the first time and completed as quickly as possible.
Javier Orensanz, GM of the Development Solutions Group at ARM, shared his insights on virtual prototyping on Thursday, June 23, during a talk at Cadence’s San Jose headquarters. ARM remains focused on two modeling paths:
ARM sees great potential in the open-source gem5, creates its own C models, and uses open-source technologies like QEMU. Customers, of course, want fast and very accurate models for prototyping, but right now, this amounts to nothing more than snake oil, said Orensanz.
Orensanz traced ARM’s history with virtual prototyping, starting with 2002, when the company’s prototyping solutions included ARMulator ISS, a simple C simulator, and the versatile testchip + FPGA approach. In 2003, the EDA industry began showing interest in modeling. The following year, ARM acquired Axys, which created cycle-approximate models. “It was a beautiful story. Unfortunately, when you looked into the details and scratched the surface…a number of problems appeared. The industry was not ready for modeling. Most people didn’t have resources to invest in virtual prototyping. The second problem was the actual effort in creating these models—most of the model creation had to be manual and it became quite expensive,” he explained.
So, from a business standpoint, things weren’t working, but the killer question from customers was: Can you guarantee that my benchmark result on the model will be within 5% of the result on silicon? “And the answer is no,” said Orensanz. “When things are inaccurate, you don’t know how inaccurate it’s going to be. The fact that we couldn’t make this guarantee of accuracy was the last nail in the coffin for this kind of business,” he said of cycle-accurate modeling. So ARM sold its SoC Designer product to Carbon Design Systems in 2008 (Incidentally, everything that goes around comes around: last year, ARM acquired Carbon Design Systems, providing it with access to 100% timing accurate models, a good library of IP models, a great library of Carbon Performance Analysis Kits™, and a scalable model creation methodology).
In 2004, ARM acquired FastSim code translation technology from Red Squirrel, and the technology evolved to become fixed virtual platforms that have been deployed to tens of thousands of developers. But for next-gen mobile frameworks, companies like Apple use their own modeling technology, while companies like Google favor open-source technology to create their own virtual platforms. It was clear that the business model needed to adapt. That’s when ARM moved on to models of hard IP, so that its customers could create their own virtual platforms and use any system simulation tools with ARM® Fast Models, accelerating their software development process.
To develop its ARMv8-A software stacks, the software development part of the process on an ARMv8-A virtual platform using Fast Models was a two-year process, while development on hardware took one day. “The time to market that we saved was huge,” said Orensanz, adding that ARM is working to extend the functionality of its Fast Models in new areas such as timing annotation, hybrid emulation, and graphics acceleration.
Looking ahead, customers will always want earlier availability of models, a challenging proposition when RTL is not complete. And customers will always want more performance. So, Orensanz concluded, it makes sense for ARM to partner with EDA companies to solve these problems. It also makes sense, he said, to establish a single standard that connects multiple modeling technologies. “The future is full of opportunities,” he said.