I. Introduction
I asked the founders of several chip companies this question.
Some say 2 years (a medium-sized circuit, MCU chip company with mature process);
some say 1 year (an AI computing power chip company )
;
It’s amazing that it’s one month (the speed of spending money...can’t help but make me shed tears of poverty); did this brother who spent one month buy an ultra-luxury architecture package produced by a certain A-letter company?
When the 100 million is burned out, in fact, many companies have not even seen a chip, and may not have a decent demo. But I'm sorry, the next financing must be credited immediately, otherwise the previous investment may be all in vain. How did the money get burned?
The chip design company does not have any particularly expensive equipment and site investment. It is an asset-light company. The main expenses come from personnel costs, tape-out costs, and various intellectual property and service fees. **These expenses, It will let you realize what it means to spend money like running water.
Take an AI processor chip used in data centers as an example. Even if it is not the industry's most sophisticated indicator, it can only barely produce a generation of products for 100 million yuan.
2. Build a team
For the chip team, first of all, it is impossible to find three architects (one for system level, two for microarchitecture IP level or for performance design and PPA), the first step is to at least set up the chip architecture Ah, according to the data in the second half of 2022, the annual salary of senior architects (architects without too little qualifications) basically starts at 150W.
Bridge Bean Sacks! This is the architect's current salary, I didn't say this is his expected salary (manually cover his face), don't forget that we are a start-up company, what should we do if we want to attract top architects to our team?
Cash, options and stocks, and vision instillation (commonly known as brainwashing or painting cakes in Jianghu) are all indispensable.
Oh, by the way, if the founder has strong persuasive ability, personal charisma and industry influence, he can make a group of powerful bigwigs choose to join with a flat salary or even a reduced salary. I voted quickly, and I call it the second core competitiveness of chip companies.
With an architect, it takes at least half a year to build it, and the specific RTL will start to be done during this period.
Chip small scale? Recruit 5 designers (logic chip design engineers) first.
Medium-sized? Recruit 10 designers first.
Super-scale? From 20, there is no upper limit. etc.
Design alone is definitely not enough. One design should be matched with two design verification engineers. Our start-up company needs to save some money. One design engineer should be paired with 1.5 verification engineers, or one designer should be paired with one experienced verification engineer. (My mother praised me as a little money-saving expert).
Even the smallest team requires 10 chip front-end engineers (5 design + 5 verification). The
current salary range of design/verification personnel currently working is roughly like this. Taking Shanghai as an example, fresh masters are generally 30 -50W; 2-5 years generally 40-80W; 5-10 years 50-100W; more than 10 years mostly start with 100W.
Regarding the salary and position of IC design, there is a detailed introduction on the website of Xinxuechang. If you want to know about IC design, you can go to Xinxuechang.com .
The more high-end chips, the more senior personnel are needed to join. In order to save money, the founders hold meetings during the day, manage, raise financing, and return to the company at night to do technical work on the front line, which is really not easy;
When RTL is about to freeze, don't stop, DFT (design for testability engineer) and PD (physical implementation engineer) personnel are about to start.
It is very expensive for a chip company to maintain a complete back-end team. Most of the back-end jobs belong to the project system, and they will only get involved before the chip tape-out node (it will be exhausted in stages), and the workload is often not saturated in the architecture and front-end stages, so many chip companies choose to outsource the work of the chip back-end To a chip design service company (design service), but the cost of a design also ranges from hundreds of thousands to several million.
Although chip companies are hardware companies, software engineers are also indispensable. Due to the fire of AI chips in the past two years, a once unpopular category "compiler software engineer" has been fired for sky-high prices in the past two years.
At the design stage of the current chip, software and algorithms have to be considered and embedded. If chip design engineers are only expensive, then chip algorithm engineers are really expensive, rare and arrogant. It is normal for a senior AI chip algorithm engineer to earn twice the annual salary of the founder.
Calculated in this way, if only the basic personnel are barely equipped or some design service companies can be found to help, about 2000W is ready to be spent. So far, it has just formed a team that just needs the version, and the chip is not even half a shadow. No.
3. Architecture
At the beginning of the business, the founding team had to confirm which instruction set architecture camp to join. Now most of the low-power logic chip companies use the ARM company’s architecture, and the companies that make mobile phone chips are basically in the ARM camp. ARM's products are good but not cheap.
Remember the luxury package bought by the big brother in front?
The starting price of ARM's architecture is usually calculated in millions, and the unit is not RMB but US dollars! Want to see the source code? Start with 10 million US dollars (Of course, we are a start-up, first buy an IP for use, and don’t want to look at the source code, even if this is the case, it will cost several million). This is only the initial authorization fee.
ARM will also charge a certain percentage of fees for each chip sold afterwards, which will be a considerable expense for chip companies in the future.
Since buying an architecture is so expensive, can I choose the free RISC-V architecture (you can understand it as the Android of the mobile phone system, open source, cheap or even free), of course!
But the follow-up ecological construction costs are ridiculously high. If you look at the companies that use the RV structure, who doesn’t have ten or eight marketing and sales, and this marketing and sales team spends millions of dollars a year. The boss figured it out, and bought ARM with tears in his eyes.
**Next, the EDA software team came oncoming,** they took solid steps and said to you: Comrade, it’s time to pay. If there are various ways to save money in terms of staffing and instruction set architecture, then in terms of EDA software, this money really cannot be saved. The design drawings of all chips must run on EDA software. You can understand that it is not the same as Adobe's Photoshop for people who do web graphic design.
Genuine EDA software is currently monopolized by several giants, whether it is from Synopsys, Cadence, or Mentor, it costs several million (US dollars) to buy a set, even if it costs tens of millions to buy The EDA simulation tool only gives a dozen licenses.
So can I use a pirated version? Unless you don't want to take the chip to tape out, you should honestly buy the genuine one.
Does the money for virtual things end here? The pattern diagram is simple.
The next step is how to solve a bunch of IPs, design by yourself or buy or find a design service company to do it. In short, no matter which one you use, you have to swipe your card.
How to solve these IPs such as HDMI, PCIE, DDR, etc. of the interface class? If you buy an IP directly, it may cost hundreds of thousands or even millions, and each license has an authorization period, and you have to renew when it expires;
Hiring engineers to do IP by yourself, the annual salary of a small team is several million. It is also reasonable to find a design service company. The owner of the outsourcing company has to make a little difference, but at least he can roll up his sleeves immediately after taking over the job. People recruited by themselves may have to work for a while, and it will take time to verify whether they are reliable or not.
If it is a SOC baseband processor chip for CPU or mobile phone, dozens of IPs are waiting for us to complete. All in all, this piece is hundreds to tens of millions of expenses, and there is no upper limit.
A rough estimate of the above expenses has already reached the price of 5000W, 100 million yuan, half of which has already been spent on these invisible and intangible things and services.
After opening a chip company, I suddenly felt that the Hermes bags of my girlfriend seemed so close to the people. But the companies that can spend the money smoothly are actually stronger than many companies. Some companies have been unable to recruit people for more than half a year, and have been hesitating strategically whether to do their own research or save something to deal with it first. Investors After reading it, I really want to breathe out the fragrance.
4. Tape-out
After preparing to start filming, the speed of spending money is like a rocket. Taping out is like taking a big exam for chip companies, and the registration fee for the exam is surprisingly expensive.
I couldn't help but interject. The scene before the chip design company's tape-out was the same as the classroom scene before the college entrance examination. There were cheering banners everywhere: the final 60-day countdown to the tape-out battle... The atmosphere was very similar to before the college entrance examination.
Generally, small companies will not choose the full mask for the first version (package a whole wafer), but will use the mpw method, similar to Pinduoduo’s mode, where several design companies share a wafer, but It is necessary to ensure that the same process is used; regardless of whether the order is made or not, the mask (Mask) needs to be paid in advance.
Because the manufacturing equipment of Mask is expensive and scary, the masks made are also invincibly expensive. The starting price of a mask plate for a 65nm process node is about 100W.
If the 28nm process is used, a set of mask plates starts at 300W, and if it is 14nm, a set starts at 500W.
Oh, by the way, generally one set is not enough, you may need to prepare 2 sets. Investors' hearts are bleeding.
Of course, the mask version belongs to the mask version, and the money for tape-out still needs to be paid separately. The price of a single stream is equivalent to burning a house, this is really no joke. A tape-out of mature technology may be a suite in Shanghai Outer Ring, a tape-out of 48nm process may be a house in Zhangjiang, a tape-out of 14nm process may be a house by the Huangpu River, and a tape-out of 5-7nm process may be a house in the French Concession of Shanghai. suite.
Speaking of which, 100 million has basically been spent.
5. Test
After the house is in hand, oh no, after the chip tape-out is in hand, it will be tested immediately. Tape-out does not mean that everything will be fine once the tape-out is successful. Usually, a chip will have at least hundreds of bugs just after tape out, and it will take a year to fix these bugs.
The team member above who burns a lot of money: Boss, our salary is about to be settled. Fixing bugs really requires racking your brains, turning so many green youths into bald uncles. If you can’t solve the problem, you have to find a professional failure analysis company to help you. Fixing a few bugs and hundreds of thousands can’t get away, right After paying the money, please queue up at the back, there are still 10 chip companies in the front.
Of course, the occurrence of bugs is not all design problems, and some of them also involve the process of the fab, and the engineers of the fab are looking for problems. I'm sorry I'm too busy, why don't you find a third-party testing agency outside to help you find process bugs? Once again hundreds of thousands were burned.
Assuming that all the bugs are solved in one year (even if there is still one bug, the chip is still just a brick!), I am going to find Changdian and Huatian to do packaging and testing. Although it will cost money, the price is cheaper than the fab Streaming is indeed a lot closer to the people.
Finally, the chip is finally ready.
For the next round of financing and corporate publicity, the exciting press conference can finally start at this time~ Holding a chip in hand, facing the camera with a smile, "Our chip, YES!"
6. Volume
If you compare running a chip company to learning from the Western Paradise, when the first generation of chips starts to increase in volume, you may have just left Chang’an City at this time, how to enter the market, how to increase the yield rate of mass production, and how to reduce costs and increase gross profit Interest rates, how to incorporate your own chips into terminal solutions, how to obtain various certifications (especially automotive chips!), and various business negotiations will be so complicated that you will doubt your life.
The circulation of inventory, the management of the supply chain, and the acquisition of production capacity, every step is on thin ice, and there are people or institutions that need to spend money to solve and manage each step.
By the way, when the first-generation chip has not yet started tape-out, the development of the second-generation chip has generally begun, and each of the above costs has to be repeated.
So for a chip start-up company, if the first tape-out fails, if the story is told well, the next round of financing will be able to support it, otherwise it is very likely that it will not be able to survive the second tape-out. up.
Today, a big guy in the industry told me: The team of a chip start-up company is going to work hard, then decline, and die three times. I deeply agree that the team cohesion is basically lost if the two tape-outs are unsuccessful. This is something that no amount of money can buy back.
Therefore, once a well-known investment manager in the circle summed up the three major methods of investing in chip design companies: asking for a lottery, measuring the horoscope, and looking at the face.
Before investing, do the fortune telling for everyone in the founding team to see if this group of people will be rich and powerful, and if not, don’t waste money. Although it is a joke, it still expresses the bitterness and difficulty of chip investment.
