Why semiconductor recovery takes time

Advanced manufacturing operations

Car production is essentially an assembly operation, albeit a complex one that Henry Ford couldn’t possibly recognise. It uses a standard bill of materials (BOM) and JIT delivery approach to enable operatives and robotics to integrate numerous components and sub-assemblies into a final car. But I’m getting ahead of myself: car manufactures first have to make significant financial investment in engineering design, modelling, prototyping, the components to use, the manufacturing operation, and finally sales and customer support.

They employ lots of people both directly and within their country’s wider supply chain and so garner a lot of political and media attention, but compared to the complexities of semiconductor production automotive manufacturing is decidedly low tech!

The most technically complex and challenging process, period!

Advanced semiconductor manufacturing starts with a silicon ‘boule’ (an object the size of a large log), produced by only a handful of organisations with the necessary infrastructure, technology and experience to melt polysilicon with boron and phosphorous at 1,420˚C. These highly specialised companies go on to ‘seed’ the boule with a silicon rod before ‘rotating and pulling’ it to produce pure polycrystalline silicon.

The ‘boule’ is then sliced into 1mm-thick ‘wafers’, which are then lapped, chemically etched and polished to exacting tolerances and either annealed, vaporised, ion implanted or bonded before being consigned to an inert-gas filled ‘carrier’ for onward ‘travel’ through the multiple stages of the particular ‘batch’ production process the semiconductor manufacturing customer employs. Typically, the first manufacturing step involves oxidisation, the application of insulating and conducting materials and a photoresist before the wafers are ‘cooked’ in a furnace at an unbelievable temperature, which is why semiconductor plants are often referred to as ‘foundries’.

The manufacturer then ‘prints’ his semiconductor schematic on to the wafers using UV light or electron beam lithography. When the photoresist is chemically removed the unprotected layers are exposed and the wafer is returned to the furnace and heated. Unprotected areas of the wafers are cleaned by gases and chemicals in an ‘etching’ process. Finally, a conductive layer is applied (known as ‘doping’) and metal is added by a chemical vapour deposition process before the wafers are ‘etched’ once again.

This ‘oxidisation and coating, lithography, baking, doping, metal deposition and etching’ cycle is repeated hundreds of times until all the transistor layers have been completely deposited on the batch of wafers. Typically, each ‘cycle’ takes twenty-four hours due to the heating/cooling required and the availability of equipment and capacity, so it’s no surprise that the lead-time to produce a batch of wafers is directly related to the time it takes to complete each process cycle and the number of cycles required.

The entire process is subject to exacting statistical process control standards to ensure the manufacturing process is reliably repeated and any process flaws identified and corrected quickly. The value of the wafer increases with each process ‘cycle’ and an error can effectively scrap a entire ‘batch’ of partially process wafers, which will then take ‘n’ cycles (time) to recover,

Once the initial ‘front end’ semiconductor wafer manufacturing process is complete the ‘batch’ of wafers are returned to their ‘carrier’ ready to undergo ‘back-end’ assembly into a physical package before undergoing multiple test cycles to establish critical performance parameters i.e. speed, power consumption, and so on., and finally inserted into tape & reel, tray or tube packaging. This is also a surprisingly complex process and there is inevitably a degree of ‘yield loss’, which only significant on-going investment in specialist production capacity can mitigate. No surprise then that this concluding stage is often outsourced to specialist third-party organisations.

Risky investment

To operate profitably a semiconductor foundry needs to continually operates in excess of 70% of its planned capacity but ideally would deliver capacity utilisation in the 95+% range with a ‘process yield’ (known good product) of 99+%. Investors need deep pockets and a healthy appetite for risk as the operating costs and capital equipment upgrades also need to be funded over the depreciation period, as do the carefully choreographed maintenance operations, and the funding costs (the interest rates charged) reflect the very high risks involved.

Conversations with civil servants and industrialists about the investment needed to support advanced semiconductor manufacturing invariably triggers wide-eyed disbelief and the assertion from these worthies that “there must be easier ways to generate growth and profit than by making such huge and risky investments”.

They could well be right: Building a state-of-the-art semiconductor foundry with the capacity to process 50,000 wafers a month is projected to cost circa $15 billion. This investment must be depreciated in three to five years, which suggests that output sales revenues – internal and/or external – need to be in the range $3-to-$5 billion per year, demanding 24/7/365 operation. The short history of the semiconductor industry is littered with organisations that haven’t achieved these targets and become financially successful, surviving primarily by merger and acquisition to bolster both the market opportunity and the economics of size, scale, learning and knowledge.

The rise of fabless semiconductors and third-party foundries

In the 1980s it was possible for an integrated device manufacturer (IDM) to own all stages of the semiconductor manufacturing process, but costs rose steeply and quickly in line with device complexity, and the cost of building a state-of-the-art semiconductor foundry also escalated. Together with the increasing number of semiconductor start-ups (and there were over 1,000 worldwide ‘starts’ by 2000) this created the need for a new breed of dedicated third-party foundry ‘partners’ able to couple high process yields with high-capacity utilisation to profitably produce semiconductor products for multiple – and often competing – customers.

The success of these dedicated ‘pure-play’ foundries resulted in a rapid migration away from IDMs towards outsourcing, but since the financial crisis in 2008 and despite reasonably strong growth forecasts, semiconductor manufacturing capacity has remained restrained with investment only made on a ‘when absolutely needed’ basis, often involving the adoption of the latest process nodes to maintain competitiveness. In recent years even dedicated third-party manufacturers have really only been able to add new and advanced semiconductor manufacturing capacity with the financial support of their customers, usually conditional on some form of output guarantee.

Flat-to-over investment

The apparent willingness of central governments to invest in semiconductor manufacturing and the recent uptick in the availability of industry funding has given rise to concerns by financial analysts and semiconductor forecasters to expect a period of overcapacity and lower prices, probably in 2023/4. In the shorter term I believe that the availability of proprietary semiconductors will ‘normalise’ at 12-to-16 weeks during 2H’22, with commodity semiconductors normalising by the end of the year.

Given the current high levels of geopolitical tension I could be a calendar quarter or so out however. In the interim my best advice is to continue to work closely with your trusted, long-term supply partners. These partners value your custom and in order to maintain the ongoing supply relationship, are highly motivated to get the products your organisation needs to you on time.