Why I think Intel is restarting its foundry business?
With coverage on Intel’s history in semiconductor manufacturing, the competition with TSMC and Samsung, and potential threats of IDM 2.0
Shocker! Intel restarts the foundry business.
Intel said it was going to shut down its custom foundry business last year. The news came after disastrous earnings posted by Intel last financial year. It led to many critiques and in some cases, few insiders who felt that Intel should no longer be in the foundry business. Hence it was surprising when the new Intel CEO Pat Gelsinger, announced Intel would not be giving up this side of its business.
Not just that, it would be going knee-deep into the foundry business with a few tweaks. Intel plans to establish a couple more foundries in Arizona, where TSMC the world’s largest manufacturer of semiconductor chips, is establishing a foundry for the first time in the USA.
A completely counter-intuitive move, Gelsinger has proved to be the ringmaster Intel needed. But is this the right decision?
The background
Intel wanted to shut up their oppressors who openly questioned its inability to scale down (to scale up) the technology that is used. When TSMC and Samsung could seamlessly move towards the 5nm technology, Intel suffered. They consider this as a payback opportunity for the years to come when the TSMC plant in Arizona will be up and running.
Before we get into the discussion, it is time to clear out the basics. A foundry is nothing but an assembly line for semiconductor technology. The chipsets in your phone, the processor in your laptop, and the transceivers in your TV are all made in foundries which are then put together in the factories of the manufacturer. The manufacturer usually gives the layout of the chip to the foundry who then translates it to a physical entity- the brain of any electronic system namely the System on Chip or SoC. Essentially two people work to get the product out- one who makes the brains of the thing and the one who assembles all the parts with the brain.
The current market
The current market is essentially a triopoly, other companies have very few holdings. The next Though there are smaller businesses, the Taiwan Semiconductor Manufacturing Company (TSMC) and Samsung control the scenario. UMC comes a little far third. TSMC has been established a long time before Samsung came into the picture. However, Intel was present since its establishment. TSMC’s sole gameplay is a pure foundry, unlike Intel’s which are integrated device manufacturers. Essentially, it means that Intel till now, before 2013 did not have any customer focus for its foundry. That idea however did not take off well. Though a few companies had agreed, Intel was not able to scale its business as quickly as its arch-competitor TSMC. Most of the chips made in the foundry were for Intel itself. They could not produce anything other than their portfolio or over the established intel’s sophisticated yet niche technologies. TSMC could quickly move from the 10nm process to the 7nm process and now (I won’t say finally) to the 5nm process. We can’t move to the Picos hopefully because of submicron lengths causing erstwhile electrical effects to come into play. Its inability to scale quickly put it out of competition. Intel’s game has not been strong for a while, due to which people have removed it from its cult status. Many people wanted Intel to give up its foundry business, which established under Andy Grove’s leadership faltered. However, Gelsinger has given the clear green flag to in fact expand its business.
Why did it all go downhill?
- Problematic business model
Intel’s business model was a little problematic. TSMC’s business model was diplomatic.
The diplomatic kid is surviving, whereas the problematic kid is trying to find its feet in the waters. Intel’s foundry came late to the market of customer — focused foundry business, but it won big with Apple.
- Inability to scale
However, due to its inability to quickly adapt to newer technologies, and also deliver on them credibly, they failed to retain the customers. But Apple decided to sever ties for Intel’s inability to scale down to 5nm. Other foundries do not shut down the other technologies which diversify their portfolio of customers. They always create a hybrid structure for the new technologies, that allow the older ones to co-exist. Not all integrated circuit systems require 5nm. In many cases, a majority of systems are created even at 22nm and above. Their performances do not require high-end manufacturing. Additionally, not everyone in the market can afford the latest technology. A single product multi-vendor marketplace is always beneficial in price setting, and competition.
Intel foundry business was drying.
IDM 2.0? What’s up Intel?
The shock of Gelsinger talking of IDM2.0 is the talk of the town right now. He has revamped the existing business model, that tries to address the current supply chain issues related to semiconductor shortage (a misnomer).
It is not the lack of silicon that is causing a semiconductor shortage. It is the lack of assembly lines that can put silicon to good use. The current firepower present in foundries is not able to meet the demands. The Covid-19 lockdowns emphasized the greater purchase of consumer electronics for home and work comforts. As the demand increased exponentially in a brief period, the gap with supply which is a constant value was also widening. The current manufacturers saw a good increase in their valuations and turn their stocks. The number of business deals getting signed was bringing in voluminous money into their barracks. This begs the question that more money, more investment, many new factories, and in turn higher returns.
Just that making semiconductor chips is not as easy as making potato chips. Setting up a fab takes around 8 billion USD. It takes an additional couple of years for the fab to operate at its maximum potential. As the process sizes are decreasing, the design cycle is taking more time to complete. A lower-end process says 22nm may take 6–8 weeks to complete, while a 10nm process may take 11–14 weeks. Nice time crunch right?
Lays’ makes 124 million bags of chips in three months. A high-capacity foundry can work at 14nm can make around 1,50,00 chips per month. It is nearly 830 times lower. The consumption is comparable nevertheless.
Gelsinger unveiled a three-step recuperation process for Intel, which was named “Integrated Device Manufacturing 2.0”. The reference press release is here.
The pros and cons of IDM 2.0
- The Pros
- The eye-catchers are the 20 billion investment on two new foundries, which would be established in Arizona, where the rest of its fabs are working. These foundries will work with the EUV(extreme ultraviolet) manufacturing technologies. This business is now an independent branch of the design business, where Intel’s promised to get them up and running by 2024. The manufacturers of foundry equipment are already in high demand. Intel can overcome the same as it has a 10% stake at ASMC, the manufacturer. Intel could potentially be on a waitlist. Intel wants these foundries to flourish to the level of becoming the biggest foundry provider for the US and EU. It is a good strategy by Intel but still comes with its cons. Now, Intel has overcome the roadblock of producing only niche products. It has already attracted the attention of many customers including, the likes of Qualcomm.
- With many countries having tense ties with China, trying to redirect most of the foundry business away from China by introducing a new foundry seems logical. To overcome the issue of over-customization, it is looking to have standardization in the design flow by partnering with EDA companies like Synopsys and Cadence. Though the world is facing a semiconductor shortage currently, it may be sometime before the operations begin at that plant before the ripple effect could reduce. However, there is still a possibility Intel may invest a part of the 20 Billion in bringing flexibility to the existing plants in Arizona. Now, it could potentially loosen up the rat race of automobile manufacturers to get their devices delivered.
- Extreme waitlists present in the foundry markets can reduce if Intel promises to deliver its chips at the right time. This strategy removes the lack of flexibility and in turn diversification that the old Intel lacked. The image value which Intel had — staying active in the competition, being a market leader, and closely trying to enter the mobile phone market will be an added advantage now. We all know how Intel faltered in the smartphone and 5G modem businesses. This could be an effort to salvage the ever-growing need for the same.
At this point, I cannot go forward without stating Mckinsey’s “The potential shake-up of semiconductor business models” which was published in 2013. It discusses Intel’s scenarios in the future. Speak of the devil, Gelsinger is essentially following at least 2 of the four listed. One of the scenarios stated is impeccable. It envisages that Intel becomes a successful foundry. I feel it is applicable right now.
Additionally, the third scenario, where fabless companies, with a need to quickly get their designs out, can invest in foundries (in this case virtually), is beneficial to both the parties involved.
- The Cons
- The plants will be established in Arizona, where TSMC has also planned to construct its facility. Samsung has decided to do it at Austin, a place quite near to Arizona. This begs the impending question of sustainability in manufacturing. Arizona is a place with lower taxation and good resources required for these facilities. However, having multiple manufacturers in the same zone can often cause supply chain issues if a calamity hits (though that case in Arizona is less). Consumption increases, people relocation increases, sustainability and the idea of decluttered workplaces (which promote sustainability) are threatened.
- Moving on to the second part, Intel has decided to expand its association with third-party foundries to produce its “tiles” (IPs to go on SoCs). I feel this though this seems counter-intuitive to the idea of an IDM having cooperation will always help Intel pump in more profits. This will enable it to expand its portfolio and work on the agenda of its revamped business model — try to get back to being the Old Intel, which was always 2–3 years ahead of the competition. Now, Intel can concentrate its skilled workforce on committing to design projects more than foundry services (ie- before its independent facilities are up, which requires a huge add-on to the workforce). I would like to call it the “tweaked version of Mckinsey’s fourth scenario” in practice. Instead of asking for their fabless customers to invest in the processes, they are technically are getting their chips ready from third-party foundries (in turn, a form of incoming investment to their current company) that can be reallocated to the foundry business.
Though Intel’s strategy now may seem fail-proof with the semiconductor shortages and Biden’s administration trying to boost up the economy, to sustain, Intel needs to execute faster and deliver on its promises. It should be flexible to change its strategies without sticking to one. Forecasting the profits, the company should make sure the customers “stick” with the company for a long time, to revive the lost profits. All this would need a good work culture that fosters trust, honesty, and patience. All in all, it is an attempt to bring Intel inside the competition and not exactly giving them the winning edge. Once the company achieves redemption, it should change the strategy to become the winner of capital markets.
It should not focus on insane cost shredding in terms of manpower. Let us hope Intel sees a ray of hope.
