SANTA CLARA, Calif.—It's a rare moment when you can see the train wreck coming and the collision-avoidance potential at the same time.
We're at a moment like that now in electronics design.
This occurred to me as I heard ARM CEO Simon Segars (pictured right, courtesy of Chris Edwards, Tech Design Forum), keynoting ARM TechCon 2013, utter an oft-heard phrase: "We are at the brink of a new era of innovation."
My first thought was "Yes, but what new era? This industry innovates new eras seemingly every single day."
Data crunch
But this was different: Segars was describing the potential benefits and challenges as we embrace the Internet of Things (IoT).
And his keynote came a day after Martin Fink, HP's CTO and director of HP Labs, delivered another address. Put the two of them together, and that's when the train wreck and the possible collision avoidance come fully into view.
The train wreck looks like this:
- Global mobile data has grown 50% this year to 900 petabytes a month, Segars said.
- Cisco says that by 2017, mobile will grow by factor of 10 to 8 zettabytes of server data.
(If you think about it, that projection is probably conservative because IoT is in its infancy and likely will explode with applications and use models in the coming months and years.)
Now shift to the Fink side of the train-wreck scenario:
- The network we have today is the network we've had for decades, handling massive amounts of transactional data.
- The systems powering those networks are based on 60-year-old computing architectures and paradigms.
- We're entering an age in which streaming data is becoming more important and our "device-centric" world will quickly change to accommodate consumer needs.
Segars put it this way:
"The network we have today just can't absorb orders and orders of magnitude more data. We're going to have to change the network. The network itself is going to have to evolve. Silicon content is going to have to grow."
There are differing evolutionary views, but a valid one is this: Companies that are fabless and leverage the flexible IP model will be able to be agile, manage costs, and meet the need for the emerging class of data center workloads, and their requirements, for optimizations.
New computing paradigm
For his part, Fink sketched out a vision of a new network and computing architecture to address these changing macro trends.
In the vision, today's constrained memory approach (temporary memory space and storage silos) is combined—a "massive universal main memory pool that removes hierarchies of data shuttling and looks at it as a single entity."
Continuing on, 20 years of general-purpose computing is supplemented with a different approach, "a move to energy- and algorithm-optimized ecosystems."
Said Fink:
"Increasingly, we will start to see the data we want to process and the processors specialized to what we want to do. Do I want to deploy a general-purpose processor ... to do video processing and video analytics? Maybe I want to create a specialized SoC that only does that."
That's where IP-based design, language-based design methodologies, and the like begin to hold sway.
Add in a shift from copper interconnects to photonics interconnect and you see a completely different approach to computing and networking that offers better performance and lower energy consumption.
Boldly go
HP's version of this vision is called Moonshot, a new server architecture that the company claims is 80% less costly, 77% smaller, uses 89% less energy, and is 97% less complex than comparable contemporary servers.
Given that development and the large and robust design ecosystem that IP and EDA vendors continue to build and evolve, you can see that that out there on the railroad tracks, there's a switch that's being thrown.
And the big innovation train will roar right along down a different wide-open track, picking up steam along the way.
Brian Fuller
Related stories:
—ARM TechCon Panel: Embedded Software Development Goes from Isolation to Collaboration