Foundry, Fabless or Fab-lite?

The high costs required for ­producing high-end ­microchips have resulted in a clear sepa­ration of chip development and semiconductor production: the majority of semiconductor ­products are now produced by purely production ­suppliers, while most semiconductor ­developers no longer have their own factories.

Anyone who works with microchips is sure to be familiar with Moore’s law. This law states that the number of transistors that fit in an integrated circuit of a specific size doubles approximately every two years. Moore‘s second law is less well known: this law states that the costs for a chip factory double every four years. This is a direct consequence of Moore’s first law, as the smaller the chips become, the more complex, challenging and therefore expensive they are to produce. If you consider that a single, state-of-the-art EUV system (EUV stands for extreme ultraviolet) now costs around 150 million US dollars, it’s no wonder that an entire chip factory quickly accounts for more than 20 billion US dollars. 

Only a few large companies can afford this. The number of factories that produce microchips has actually continuously decreased over the years. At the start of the semi­conductor era in the 1970s, all leading semiconductor manufacturers tested and produced their microchips all by themselves.

Production outsourced to foundries

An increasing number of smaller manufacturers entered the market in the early 1980s; these companies developed chip solutions for very specific applications. However, they did not have the necessary resources to produce their chips themselves. In contrast, the large IDMs (integrated device manufacturers, i.e. companies that develop and produce their own chips) had the production capacity and were happy to be able to take advantage of the high utilisation of their factories to reduce the costs per chip produced. They therefore started to produce semiconductors for smaller “factory-less” or “fabless” companies. However, since the IDMs primarily produced their own microprocessors and semiconductors, production for fabless companies was often treated as less important. This changed in 1987 when TSMC set up the first factory to produce products solely for fabless customers. This laid the foundation for the new business model of “foundries”. Foundries can completely focus on optimising production engineering and achieving high factory utilisation. This is particularly important since, according to calculations from the market analysts at Beroe, the fixed costs – i.e. costs independent of the number of products produced – are almost a third of the total costs for semiconductor factories. High factory utili­sation is therefore essential for ensuring cost-effective operation. According to the ­Semiconductor Industry Association (SIA), semiconductor factories work with a minimum utilisation rate of 80 per cent; at high-end factories, a utilisation rate below 90 per cent can be critical. The fab­less model has proven to be a true win-win cooperation: the foundries can achieve a high, stable utilisation rate. The fabless companies save on the high investment and
operating costs required to have their own manufacturing plant and equipment. 

Limits of the fabless concept

This business model is currently increasingly reaching its limits: on the one hand because the most state-of-the-art semiconductor products (currently with a 5-nanometre structure size and soon to be with a 2-nanometre size) can now only be produced by three foundries globally. On the other hand because the foundries are establishing new production capacity primarily for these semiconductor technologies – the profit margin is simply the greatest here. However, these high-end chips are simply not necessary for many fields. Of the almost 1000 microchips currently installed in a state-of-the-art car on average, around two thirds have a structure size of 130 nanometres or more according to ZVEI. Combined with the relatively low quantity that the automotive industry commissions foundries with and against the backdrop of rapidly increasing demand for microchips globally, this makes it difficult for automotive manufacturers to obtain sufficient numbers of chips.

Cheap inhouse chip production

A semiconductor business model that is specifically tailored to the requirements of an industry such as the automotive industry is increasingly being discussed. Some chip buyers are increasingly being overlooked compared to large IT and smartphone manufacturers when it comes to reserving production capacity: the fab-lite concept, however, enables these buyers to secure their supply of chips. The fab-lite model focusses on inhouse production of “bread and butter” chips, i.e. microchips whose design no longer corresponds to the state of the art, but will still work well for a number of years in many applications. This type of inhouse chip production is not only attractive for the automotive industry, but aerospace or defence companies could also benefit from this concept. Fab-lite doesn’t necessarily mean building a new factory – purchasing older factories from a foundry is also an option. For example, Volkswagen could purchase a semiconductor production plant that was no longer of interest for a foundry, but still produces semiconductor products that are perfectly tailored to the automotive manufacturer’s needs.