Intel Aims Beyond Digital Logic
Intel’s 22FFL process takes the company into new territory, with a process that is more distinct from high-performance logic nodes than ever before. Previously, the SoC processes were clear variants of existing high-performance processes. The SoC node would have different transistor formation flows, additional transistor types, more interconnect options, and a richer set of passive devices. For example, the 14nm SoC process supports multiple transistor pitches (70nm for high-speed logic and 84nm for ULP) and introduces a high-density single-fin SRAM cell. However, the interconnect stack for 14nm SoC is mostly re-used from the high-performance node, eliminating the 56nm M3 layer and enabling additional metal layers of many of the existing pitches (e.g., 52nm minimum pitch and 80nm pitch).
The 22FFL process starts with building blocks from the 22nm, 14nm, and 10nm process flows and adds in unique features such as a 4µm pitch metal layer. The new process integrates them in a novel way that is much better suited for analog, RF, and foundry-like applications. In a sense, Intel’s 22FFL team was given many more degrees of freedom and as a result has been able to develop a process that is flexible for third-party customers and compelling for low-power. While the 22FFL process is more cost-effective than prior efforts, the actual cost to customers is also determined by Intel’s manufacturing operations. Intel favors smaller fabs than TSMC, and the total cost is unclear.
The 22FFL process demonstrates that Intel can excel outside the realm of manufacturing high-performance digital logic. However to be relevant, that excellence must ultimately translate into products. Intel has not disclosed any plans for products manufactured on 22FFL, but the marketing has shifted over time. Initial disclosures focused on 22FFL as a process for mobile and IoT, while later documents emphasize mobile and RF. Technically, it could be a good fit for wireless modems (e.g., 4G, WiFi, or BlueTooth) and potentially even PC chipsets. Intel’s custom foundry is another source of customers. The foundry business is still struggling due to Intel’s 10nm delays, and it is unclear which (if any) customers will use 22FFL. However, the process is ready – the PDK was release in late 2017, and with any luck, products will be coming soon.