Moore's Law, named after Intel (INTC 0.01%) co-founder Gordon Moore, claims that the number of transistors in the same area of silicon can double every two years. Moore's original prediction, made in 1965, suggested that the number of transistors would double every year, but it was scaled back to a two-year cadence in 1975.
Source: Pixabay.
That updated "law" has held true since then, and countless sci-fi stories have suggested that it could lead to a "singularity" of self-aware computers. But during Intel's second-quarter conference call, CEO Brian Krzanich announced that the company would miss that two-year cadence for the second time in row. Krzanich also declared that further delays should be expected, noting that the "cadence today is closer to two and a half years than two."
That surprising statement fueled speculation that Moore's Law could eventually end. If that happens, what will happen to Intel, which has built its reputation on shrinking its circuitry faster than its competitors?
The business of "tick-tock"
Every Intel investor should be familiar with Intel's "tick-tock" schedule. In a "tick" launch, the existing microarchitecture is replicated with an improved manufacturing process. When Intel shrunk the 22nm Haswell to a 14nm Broadwell design and added new instructions, that was a "tick" launch.
In a "tock" launch, a new microarchitecture is introduced, but the size remains the same. That's the case for Intel's new Skylake chips, which are fabricated with the same 14nm process as Broadwell but equipped with more advanced circuitry.
When Intel was manufacturing 32nm chips in 2010 and 2011, the gap between the "tick" and "tock" was precisely one year. But the gap stretched to 13 months with 22nm chips in 2012. With the move to 14nm chips, the gap between the "tock" (Haswell) and the following "tick" (Broadwell) expanded further to 15 months. Skylake, a brand new "tock" architecture, is expected to arrive by August, which would put it roughly on schedule.
A messy road ahead
But looking ahead, things get messy. Cannonlake, the 10nm "tick" for Skylake, ran into manufacturing issues which will delay its launch to 2017. In its place, Intel will launch a second "tock" next year with Kaby Lake, which is expected to be a third 14nm chip after Broadwell and Skylake. This strongly indicates that the current definition of Moore's Law won't hold for much longer.
Meanwhile, cramming more transistors onto a CPU is producing diminishing returns in performance. According to industry benchmarks, the upgrade from Sandy Bridge to Haswell resulted in performance boosts of 10% to 15%, depending on the application used. The upgrade from Haswell to Skylake is expected to result in a mere 6.7% boost.
To make matters worse, the PC upgrade cycle has expanded from a two to three year cycle to a "much longer one," according to ZDNet's Adrian Kingsley-Hughes, due to "more powerful and more reliable PCs that have a longer lifespan." When we combine that with the cannibalization of the PC market by tablets, it's easy to see why PC shipments fell nearly 10% annually in the second quarter, according to Gartner.
The performance plateau
Intel has hit this kind of performance plateau before. Ten years ago, it found that it couldn't boost the speed of single-core CPUs anymore due to heat problems.
To compensate, Intel introduced multi-core CPUs, which are now the industry standard. It also squeezed out more performance with Turbo Boost, which shut off idle cores and increased the speed of one for optimal single-core performance. But now multi-core CPUs are running out of room to grow, and Intel must figure out how to provide meaningful boosts in performance without using expensive and impractical water-cooling solutions.
Source: Pixabay.
But even if Intel's progress in shrinking circuitry slows down, it remains ahead of its top rivals in the foundry business: TSMC, Samsung, and GlobalFoundries. Krzanich expects Intel's lead over those rivals to remain "roughly the same," meaning that it will hit the "node" (nanometer) milestones before the competition.
Intel is confident that it can maintain that status quo, thanks to its massive budget for R&D, mergers, and acquisitions. For fiscal 2015, Intel plans to allocate nearly $20 billion for those purposes alone. By comparison, Samsung recently budgeted $14 billion to expand its chipmaking facilities. TSMC plans to spend up to $11 billion on capital expenditures this year, with $2.2 billion dedicated to R&D.
The key takeaway
The potential end of Moore's Law certainly won't break Intel. The company's deep pockets and dominant shares of the PC and server markets will ensure that it remains the default choice for top-tier processors for the foreseeable future.
Instead, investors should focus on how the "tick-tock" slowdown relates to longer PC upgrade cycles and declining PC sales, which are both dampening demand for Intel's processors. This could force Intel to dedicate more of its R&D budget to newer business units like mobile chips and Internet of Things modules, which are both tiny from a revenue perspective compared to its core PC and Data Center businesses.
