[INTC] Intel Corp.
Related posts:
Intel’s audacious shift from memory chips to CPUs, as recounted by former CEO Andy Grove in Only the Paranoid Survive, is widely celebrated. During the early 80s, Japanese chip companies – owned by large electronics enterprises that cross-owned banks that supplied cheap capital – built enormous fabs, manufactured efficiently (kaizen), and came to possess unit cost advantages with which to underprice Intel and win share. With Intel hemorrhaging cash as the memory market grew, the enfeebled company abandoned memory and reorganized around CPUs. It came to dominate the microprocessor market for years thanks at first to its alliance with IBM and then to its considerable scale. In the ’80s, Intel formalized a manufacturing method, “Copy Exactly!”, where, as the name suggests, every fab was geared with the same equipment, the same maintenance procedures, even the same barometric pressure and room lighting to ensure extreme uniformity in yields.
Design and manufacturing were synchronized to a degree possible only when the two functions co-exist under a single entity. The arrangement enabled Intel’s vaunted Tick-Tock cadence of innovation, where each new process node (“tick”) was implemented first on a mature microarchitecture, like so:
By mutually optimizing manufacturing and design, Intel could customize transistors and tailor IP to meet power/performance requirements specific to Intel’s products (for instance, the company timed its High-k metal gate technology to coincide with the production of its low-power Atom processor). Merchant fabs had to meet the requirements of many IC designers to maximize throughput and realize scale economies, forcing each fabless designer to accept trade-offs that an integrated device manufacturer like Intel didn’t have to.
Not only was Intel faster at shrinking transistors than anyone else, but it solved leakage problems attending smaller nodes with innovations like strained silicon (moving silicon atoms further apart to ease the flow of electrons through transistors) and High-k metal oxide gates (replacing silicon dioxide transistor gates with a thicker material), and novel geometric configurations like FinFet (wrapping the transistor gate around a 3D conductive channel), breakthroughs that Intel pioneered and uniquely possessed for some years before competitors caught up. (the latest in transistor architecture, succeeding FinFET, is “gate-all-around” (GAA), where the gate touches all sides of the stacked channels through which electrons flow, further limiting current leakage and resulting in faster switching speeds)
(fyi: “process node” metrics – 65nm, 40nm, 28nm, 14nm, etc. – have historically referred to the length of a transistor gate, a barrier that opens when voltage is applied to it, allowing electrons to flow from one side to the other, from the “source” to the “drain”. When this number is sufficiently small, e.g. ~30nm, applying voltage to the transistor gate can lower this barrier, causing electrons to flow from source to drain even when no voltage is applied. That’s what I mean by “current leakage”)
The compute ecosystem coalesced around Intel because Intel’s unmatched silicon technology and manufacturing processes yielded predictable and consistent progress. Those advantages were complemented by Intel’s brilliant “Intel Inside” marketing program, where Intel supplied “market development funds” to subsidize OEM ads that prominently displayed the Intel logo. Throughout the ’90s, the x86 architecture on which Intel’s CPUs were designed became industry standard in chip design, as Microsoft Windows became the standard PC operating system. Hence the “Wintel” monopoly, where developers had little choice but to ensure their programs were compatible with Windows and x86. Intel then extended its PC supremacy to data center servers before then also building a monopoly in enterprise storage systems1 and a significant presence in networking equipment during the 2000s. In each domain, Intel’s architecture replaced a varied hodgepodge of fixed function chips. By the time Intel brought x86 to mobile, there were over 10mn developers programming for that instruction set. Given their ascendency in PCs and servers, it seemed perfectly reasonable to believe Intel could conquer mobile too.