Chips need to be designed for a new class of small-cell basestations
That was the message Marcus Weldon, chief technology officer at Alcatel-Lucent, delivered to several hundred processor designers at the annual Hot Chips conference.
“We are in a radical transition to a capacity-driven world served by Wi-Fi and small cells and away from a coverage world” using macro cellular basestations, Weldon said. “All the new chip set designs will have a focus on metro cells with a smaller emphasis on macro cells,” he said.
“Small cells are the only way to get 80-fold factor in additional capacity that will be needed,” he added.
Weldon predicted as many as 500 million tablet computers will be sold each year by 2017. They will, along with the rise in smartphones, drive the 80-fold increase in mobile data next-generation wireless networks will need to serve.
The needs are rising at a time when carrier’s data revenues are relatively flat and their voice revenues are falling. He suggested carriers will look for ways to charge per connection and per service, especially for new ways of shifting into the digital realm traditional “analog” jobs in banking, health care and other domains.
“We are in a staring match between content and network people where no one has figured out the model to build the network with the money in the pot,” said Weldon. “It’s the entire digital economy wallet we should look at,” he said.
As for the chips, Weldon said basestation makers such as Alcatel-Lucent are transitioning from a mix of FPGAs, DSPs and multicore CPUs to a new class of SoCs from companies such as Cavium, Freescale, LSI and Texas Instruments. The SoCs integrate DSP, CPU and accelerator cores reducing as much as four-fold the number of chips on a card and amount of power the cards require.
“In future, we see a move toward SoCs using more general-purpose processor cores for Layer 2 and 3 jobs with some hardware accelerators for Layer 1” to better handle dynamic shifts in workloads, Weldon said.
“There’s still a debate in the industry about how much pooling of baseband and core processing you can do and get reasonable performance,” he said. “The core model is changing assumptions about what to do in real time and how much can be non- deterministic—this is something people are still working through,” he added.
Telecom OEMs will need millions of the chips which could sell for $50-$100 each because the small cells will cost significantly more than $1,000, said Weldon in an informal discussion before his keynote. The chips need to support 3G, 4G and Wi-Fi and ideally should have some media processing capabilities to support content caching at the edge of the network, he said.