Fabless OEMs are developing revolutionary ways to make semiconductor chips faster and smaller. Yet this drive for specialized, hyper-powerful miniaturized electronics presents challenges on the production end for OSAT business models.
Profit and efficiency go hand-in-hand, which is why producing items in bulk is generally the most profitable business strategy for manufacturing companies. Outsourced semiconductor assembly and test (OSAT) vendors are no different, typically producing organic substrate packaging for integrated circuits (ICs) based on low-mix, high-volume business models. The problem with this approach is that it doesn’t fully meet the needs of the current market in a rapidly evolving technological landscape. With a global organic substrate packaging material market expected to reach $18.2 billion by 2027, it’s critical to understand the driving motivation behind miniaturization and how OEMs and OSATs are responding.
It’s all relative: shrink the components to shrink the whole
Organic substrate packaging materials are utilized on printed circuit boards (PCBs). PCBs are at the heart of modern electronics, and have been optimized from the original through hole technology (THT) to surface mount technology (SMT) that reduced the size of PCB assemblies with significantly higher circuit densities. The latest SMTs use 3D printers to create multi-layer, high-density interconnect (HDI) PCBs to achieve dense circuitry compositions, some having as many as 36 layers. The HDI PCB market is expected to reach $16.4 billion by 2025; yet another reason for low-mix high-volume production.
But if PCBs are the heart of modern electronics, transistors are their lifeblood — and that’s where technology advances become more individualized and the need for high-mix low-volume emerges on the OEM side.
The challenge: Getting more out of Moore’s Law
At the heart of miniaturization is a SWaP-based challenge the semiconductor industry somewhat unintentionally set for itself in 1965 that has driven the technological boom defining modern times. Moore’s Law, the theory casually postulated by Intel co-founder Gordon Moore, famously forecasted that the number of transistors in microchips will double every two years. At some point, this will no longer be possible due to the physical properties of transistors, but for now, the theory holds true.
The original transistor produced by Bell Labs in 1947 was large enough for manual applications. By 2012, Intel could fit over 100 million 22-nanometer transistors onto the head of a pin. The current industry leader, Taiwan Semiconductor Manufacturing Company (TSMC), plans to reduce this to just 2 nanometers — but it’s a race to the finish line with Intel aiming to have 1.8-nanometer transistors also manufacturing ready by 2024. For reference, silicon has an atomic size of 0.2 nanometers, meaning a 2-nanometer transistor is just 10 atoms wide.
The major limiting factor in making transistors even smaller is the physical nature of ICs. In an electronic circuit, electrons move through copper wires while interacting with the particles within the wire. This produces heat, which causes energy loss in the system, and leads to a slower velocity. The more transistors packed into an IC, the more heat it produces. At some point, keeping the transistors at a cool enough temperature to function will require more energy than the voltage they conduct.
But there’s a solution: photonic integrated circuits (PICs).
Photons don’t need to travel through wires. Their position can be used as information carriers, guided nearly frictionlessly with mirrors and lenses so they don’t lose energy through excess heat production. And it doesn’t get faster than the speed of light: photos outpace electrons twenty times in microchips.
There’s opportunity in customization
Besides HDI PCBs and PICs, fabless OEMs have other tools in their belt to keep Moore’s law alive, such as new transistor designs and 3D architectures. But there are other opportunities as well. Changes in hardware architecture have progressed faster than the codes that run on them. For example, running more efficient code with C requires fewer processes than Python to perform the same tasks, in some cases yielding computationally-heavy responses 47 times faster than the original code.
And while there are some general solutions like these that can be applied to any computer, it’s not necessarily economically in the best interests of producers or consumers. High-volume production levels have kept the steadily rising cost-to-benefit ratio of minimization research manageable, but the general consumer doesn’t need expensive quantum photonic processors.
As such, OEMs need high-mix low-volume production to innovate small batches of advanced, customized solutions to drive AI and deep learning forward. The lack of versatility has led to speculation about the decline of computers as general-purpose technology, but perhaps OEMs and OSATs can strike a balance between the economic benefits of mass production and the advancement of technology through customized solutions.
Read more: