Intel's recent troubles with its tick-tock semiconductor development cycle are not the temporary hiccough the company had downplayed them to be, with its latest financial filing admitting that it is moving to a new development strategy.
Since the semiconductor industry adopted Moore's Law, the observation by the eponymous Intel co-founder that the number of transistors on a latest-generation semiconductor tends to double every 18 months, Intel has stuck to a rigid annual development cycle known as tick-tock. During the tick portion of the cycle, the company takes its existing processor microarchitecture and migrates it to a smaller process node in order to go on doubling the number of now-smaller transistors; in tock, the company works on improving the processor's microarchitecture while retaining the same process node.
In recent years, however, Intel has found the laws of physics getting in the way of its roadmap. The company's most recent process node shrinkages have encountered delays as the difficulties of sub-10nm physics raise their ugly heads, and while Intel has been quick to downplay the issues as simple bumps in the roadmap its latest 10-K financial report tells a different story: it's abandoning tick-tock in favour of a longer three-phase cycle dubbed process-architecture-optimisation.
In the filing, first spotted by financial site The Motley Fool
, Intel explained that the shift will not affect its releasing of new products each year. What it will do, however, is mean the company's products stay on the same process node for longer: instead of a new node every two years, the target is now a new node every three or more.
Under the new development cycle, Intel would launch products on a new process node in year one, develop a new microarchitecture on the same process node in year two, then optimise that microarchitecture in year three before moving to the next process node in the chain. On the surfacve, this means a 50% increase in the longevity of a given process node over the old tick-tock cycle; with no firm timescales attached to each part of the new cycle, however, it gives Intel breathing room to spend four or more years on a given node without having to confess to a roadmap miss.