AMD CTO Mark Papermaster has said that the planned shift to 7nm semiconductor nodes is is one of the toughest process moves in several generations.
Speaking to the EE Times, Papermaster said that, while AMD planned to run its second and third generation Zen architecture x86 microprocessors on 7nm, it would likely be a 'long node', like the 28nm process, "and when you have a long node it lets the design team focus on micro-architecture and systems solutions", rather than simply redesigning standard ‘blocks'.
In addition to new CAD tools and architectural changes, AMD has found that it requires changes in the way that transistors are connected and deeper partnerships with foundries. "In 7nm, it requires even deeper cooperation [because] we have quad patterning on certain critical levels [where] you need almost perfect communications between the design teams," Papermaster said.
"In 7nm, it requires even deeper cooperation [because] we have quad patterning on certain critical levels [where] you need almost perfect communications between the design teams," Papermaster said.
Papermaster continued that foundries would likely introduce 'extreme ultraviolet lithography' from 2019 to reduce the need for quad patterning. This "could bring a substantial reduction in total masks and thus lower costs and shorten cycle time for new designs," he said.
Both designers and foundries are exploring ways to both shift to 7nm, while cutting costs, he continued. AMD and Nvidia, for example, are exploring "2.5-D chip stacks", a technique that "connects processors and memory stacks side-by-side on fast silicon interposers", according to EE Times, which notes that it's still an expensive technique.
It adds: "Apple and others are combining mobile application processors with memory in wafer-level ‘fan-out packages'. The so-called ‘2.1-D technology' is not yet suitable for more powerful desktop and server processors, but versions could be ready in two or three years". That's according to Papermaster.
He also claimed that semiconductor designers and manufacturer were achieving new density advantages at each node, and accruing cost advantages as those nodes matured, "but mask costs are going up and chip frequencies are not going up, so how we put solutions together is critical to sustaining the pace of development," he said.
Papermaster called on software developers to start making better use of the multiple cores and parallel threads on offer in order for users to gain the full benefits of current and future microprocessors - because clock speeds are not going to be increasing by much, regardless of process.
In order to compete effectively against both Intel and Nvidia, AMD has taken a more modular approach in order for circuits to be reused across CPU, GPU and semi-custom designs, said Papermaster. "We couldn't just throw hundreds of designers at a problem," he said.
The manufacturing gap between AMD - and its principal foundry partner Globalfoundries - is also being fast reduced, helping to slash the performance gap between AMD (and others) and Intel. AMD also uses TSMC to manufacture its graphics microprocessors. µ
C3-PO, R2-D2, BB-8 and other Androids
Helpful cyber vigilante gets short changed by customer services
...you know, now it's less confusing...
Firm will no longer provide updates for its first Android mobe