The Mount Everest of Capital Investment: Understanding the huge barrier of entry in the semiconductor fabrication industry

The economy drives the market. Where there is demand, there will be someone supplying that demand. Industries exist to turn a profit due to that demand. The electronics industry is huge with so many competitors on every level, with so many products ranging from heavy appliances like air conditioners, refrigerators, washing machines, to small and sleek gadgets like smartphones, earbuds, smart watches and the like. All these electronic goods have a never ending demand, so much so that the industry is facing issues in meeting it. All of these appliances have one thing in common: they all run on semiconductors.

"Well just make more semiconductor manufacturing plants", will be your first thought; if only it was that simple. The semiconductor industry manages to find itself in quite a unique position, the ecosystem surrounding it is a tremendous interdisciplinary and international achievement. The raw materials and equipment required for running one fabrication plant on a commercial level need co-operation from multiple companies across several countries.

India has some history with semiconductor manufacturing due to one government funded plant in Mohali called Semiconductor Laboratory. It is the only plant in India that has the capability to produce electronics on a bulk level. Notice how I did not say commercial, it is not built for that purpose. Mainly ISRO uses it for electronics it deems too confidential for outsourced manufacturing. Tata Electronics is partnering with PSMC to construct the first private fabrication facility in India. With an investment of INR 91,000 crore (10 billion USD), it aims to manufacture devices in the 28 nm node. Tata has just entered an industry that is comparable in importance and capital costs to nuclear power plants or oil refineries.

I would like to try and break down the costs for this plant. I am not in any way affiliated to Tata (wish I was), this is just an outsider’s perspective to where that initial investment may go. The manufacturing tools will take up the bulk of the cost. The basic fabrication processes are: deposition, doping, lithography, etching, and chemical mechanical polishing. A fab of this size can be compared to GlobalFoundries' Fab 1 in Dresden, Germany. They work on the same nodes that Tata's fab will work on. Fab 1 has the capability to process 40,000 to 60,000 wafers per month, so around 1300 to 2000 wafers per day.

Deposition machines are used to create structures on top of the wafers. Deposition of compounds like silicon nitride or silicon dioxide as insulators and metals to make electrical contacts require different machines. For very thin initial silicon dioxide layers, an oxidation furnace is used. JTEKT VF-5900 Vertical Diffusion Furnace is one example. It has the capability to process 100 wafers in one batch. To meet our estimation of 2000 wafers, we will need 20 of these machines. Now, when you go to their site you will not find a price; they want you to contact them to get a quote. Making an educated guess, let us put the figure at a ballpark of 40 crore INR (4.2 million USD) for one machine. Throwing in a bulk purchase discount of maybe 15 percent, we round the final price to 680 crore INR (72 million USD). This is a very conservative estimate. Transport and installation costs may balloon the price further. 

Figure 1: JTEKT VF-5900 Vertical Diffusion Furnace 

(Reference: JTEKT)

There are around 19 million companies registered in India. As the linked article states, capital goes where it feels safe, and the semiconductor industry is not safe at all, and yet, it is easily one of the most in-demand and lucrative prospects in present times. Out of those 19 million, only around 1000 or 2000 companies may make a profit greater than the amount it costs to buy 20 oxidation furnaces. This process is just one of many to come. Let that sink in.

Moving on to lithography, 28 nm node requires deep ultraviolet (DUV) wavelength. 28 nm is a mature node, this means it has been well researched and is used in cases requiring one focused application. Our laptops are designed to be able to do a variety of tasks we throw at it but the processor in an air conditioner does not require the same kind of processing power. DUV machines have multiple vendors including ASML, Nikon, and Canon. ASML's TWINSCAN NXT:2000i is considered the industry standard for immersion DUV lithography. It has a throughput of more than 275 wafers per hour. This seems like a very high number, but lithography is a step that has to be done repeatedly. It is the rate limiting step in a fab facility. A finished wafer will have undergone about 40-50 lithography steps. One wafer may have to undergo 2-3 lithography steps in a day. We have to use our powers of deduction for this one too because there is no listed price. 

Let us take the estimate of 650 crore INR (70 million USD) for one machine. Taking a realistic per day instead of 275 x 24 = 6600 wafers per day, approximately 5000 wafers can be processed if they had to undergo one step. Taking the upper limit of 3 times, the number goes down to around 1600 per day if we consider all those wafers undergoing the same step together, which is not true since one batch may be on step 1 and another may be on step 29. Around 5 immersion lithography tools will be required. So final cost will be 3250 crore INR (350 million USD).

Figure 2: ASML TWINSCAN NXT:2000i 

(Reference: ASML X)

I am sweating after seeing such large amounts of cash being thrown around. This is still just the beginning of the mountain that a firm has to climb to just establish a facility. Working in nanoscale requires a very sanitized and exceptionally clean room. So, the name of this special room was given as 'clean room' (a child named this I tell you). A clean room is a highly controlled environment where everything from the air to the vibrations on the ground are controlled. Precision of this level does not come cheap. According to semiconductorx, a 28nm node fab will cost around 2-6 billion USD (18,000-56000 crore INR) in clean room construction. A VLSI fab would have costed even more. Suddenly all the alarms in your brain might be ringing. All the previous tool costs would accumulate into a very large number and if we sat here calculating all the processes and their respective tool costs, we all would qualify as accountants. The construction of a clean room requires the best civil engineers and high quality materials. A small mistake can cause a humongous time and capital loss.

We have calculated for a final cost of 680+3250+25000 = 28930 crore INR. Around 31.7% of the initial investment cost. Amazing isn't it.

If setting up a plant has so many intricacies and pitfalls, when will they generate a profit? The fab is first setup, then all the tools undergo qualification tests to create smooth process flow, and finally yield is ramped up. This entire process takes 5-7 years. During this time revenue is very low, so one can forget about any profit. This is the single biggest reason constraining any player other than already established giants from investing in any meaningful capacity in new fabrication facilities. In an era where companies chase short-term profits for their quarterly reports, how is anyone supposed to find patience?

Going back to estimating how many companies can pull off front-end fabrication, the number of firms with a profit greater than 3 billion USD are the big household names. Reliance, ICICI, HDFC, all the usual suspects huddle around this very exclusive achievement. The gap between a small firm and an industry titan is the same as the distance between the moon and the Earth.

Go back to main page