Automation in Semiconductor Manufacturing IEDM, San Francisco, 1982 Keynote Speech Commentary Alongside ISSCC, IEDM is the most traditional academic conference in the semiconductor field, and it is held in December every year, centering on the device technology and manufacturing process. The keynote speeches are given on the first day, and it is customary for three people selected from North America, Europe, and Asia to deliver speeches. In this year, IBM in US, Siemens in Germany, and Hitachi in Japan were selected. In 1981, Japan was at the top of the world in the memory technology, and a lot of attention was gathered in Japanese manufacturing technology. I think that the turn has come around for Hitachi, since Hitachi was a leader in advanced devices of 3 micron process, such as 64K DRAM and 16K SRAM. The paper on the Technical Digest was jointly written with Hiroto Nagatomo, Manager of Production Engineering Department, and I made the presentation. It was my first experience to make a speech at a large overseas academic conference with over 2,000 audience, and it was not easy to keep talking without letting the audience get tired for an hour. It is most important that the content itself is fulfilling, but it is also necessary to incorporate some appropriate humor and to moderate the atmosphere of the venue. The first and the last slides were prepared with such intention, and I took in the taste that would invite laughter even in the middle. I wish it would help as a reference for your speech. As a result, this keynote speech was accepted very favorably. After the conference, Michael Adler, Program Chairman, sent me the following thank you note: "--- In particular, as to your speech, many people gave me comments that it was the best one in the past IEDM speeches. Not only was the content excellent, but also the presentation with humor was very well accepted. A lot of people were surprised at the depth of your understanding about American humor ---- " This is not simply a compliment to my speech, but I think that it was the high evaluation given to Japanese semiconductor technology that had reached the world's top. Continue to slides 1
Keynote speech at IEDM in 1982. I talked as a representative of the Asian region. It was a comprehensive speech on the automation of manufacturing advanced devices. IBM person talked on Device Technology for LSI, and Siemens person on Power Device. 2
I started with a slide that took the audience by surprise, and it roused the laughter and applause in the hall. "Today's story is the same as a chef's story about how delicious dishes can be made. Let me talk about various kitchen utensils and how to use them. However, the secrete treasure of traditional family seasoning is not included! 3
4
Technology in the semiconductor field has several characteristics. We must keep these points in mind prior to the implementation of automation. 5
Advances in semiconductor technology are extremely fast, and new waves rise with the period of about seven years. Beginning with the Ge transistor, going through Si transistors, ICs and LSIs, VLSI started to rise. VLSI will be the biggest segment in the 1990s. In promoting the automation of manufacturing, technological changes have to be taken into considerations. 6
Starting with 1K DRAM, and going through 4K and 16K, 64K DRAM is starting to rise. However, within the next few years the new generation 256 Kbit will surpass 64 Kbit. 7
Although the number of bits grows rapidly, the growth in the value and quantity is moderate. This is because the cost per bit keeps falling, and the integration density keeps going up. This point has also to be considered in manufacturing automation. 8
The trend of complexity, feature size, and chip area are shown from 1 K to 256 K DRAM. The integration level increases 4 times every 3 years, the pattern size is reduced to 70% (the density is doubled) in 3 years, and the chip area is doubled in 3 years. 9
The trend of new technologies introduced after 1970. All these technologies were "innovative" at the time of introduction. 10
Actually this slide was also the one that invited big laughter in the hall. "Everyone, this shows pictures of memory cells after 16 K. It is getting smaller and smaller in every generation. The 1 Mbit cell is so small that you probably cannot see it from your seat. " 11
The following three points are challenges for automation. 1) decrease in yield, 2) decrease in throughput, 3) increase in investment amount 12
13
14
The basis of the wafer processing is that the following three processes are repeated. 1) diffusion / implantation, 2) photolithography, 3) deposition (CVD, metallization) 15
16
17
18
The wafer fab has class 1,000 clean process area in the center, and maintenance area of class 10,000 on the outside. Wafer transport between work areas is done by a computer controlled transport system. 19
Processing equipment are placed on the left and right side of the work area, and an automatic transfer system (robot in the middle of the right side) carries wafers between work areas. Every one, I think you are strongly impressed that there is not a single person in this automated line." After a moment of silence, "In fact, this picture was taken in the factory on a holiday. A big laughter followed. 20
The processes in the work zone are done by an integrated production system. 21
While wafers are loaded from the left side, all processes are performed during one round trip and unloaded from the left side. Through the whole process, the wafer is never touched by human hands. 22
23
No human intervention from loading to unloading. 24
25
26
Compared with manual work, chipping at the wafer edge is reduced to 1/10 in the case of automation. This is an example that automation contributes to quality improvement. 27
A robot that automatically carries wafers. While traveling on the orbit, it checks the obstacles etc. with the two eyes in front. 28
With the introduction of automation, manual handling decreased to 1/5 in the decade from the early 1970s to the early 1980s. 29
Compared with the conventional method, the productivity is improved by 1.5 times with the automated method. 30
This shows that the number of foreign particles on the wafer has been reduced by automation from 70 30 10 5, from month to month. 31
Several slides following this are related to the yield of LSI, which was the theme which I was good at. This figure expresses the defect distribution on the wafer. There is a region (G) in which good chips can be possibly obtained, inside the area (W) of the entire wafer, and the ratio G / W = AUF is called an area usage factor. It is assumed that the defect distribution in G is Poisson distribution (stochastic distribution). 32
This is a formula used as a basis for yield analysis at the time, and even today the basic point is still valid. The yield in the region G decreases exponentially as the product of the chip area A and the defect density D (A*D) increases. The total yield Y is the product of the area usage factor AUF and the yield in the region G. 33
When a new product is launched, the yield is initially low, and then it improves along the learning curve. For example, in the case of a memory, it may be imaged as 1 K, 4 K, 16 K, and 64 K DRAM. 34
This shows that the defect density has been reduced by about an order of magnitude in the past 10 years. It is predicted that this trend will be accelerated further. 35
Types of yield improvement are classified into three typical types. A is a case where AUF is high while there are many defects in G, C is a case where there are few defects in G but AUF is low. B is in the middle. You can see the effect of improvement by plotting the path of yield improvement. It is important for all team members to share this figure because the yield improvement is often advanced by the team activities. 36
This shows the pace of improvement in defect density in conventional manufacturing lines and automated lines. It shows that the reduction speed of the automated line is nearly twice as fast. This material emphasizes that the effect of automation is not only labor-saving, but also the effect of quality improvement is great. 37
38
39
40
The back-end process flow is shown from the probe test to the final test. The biggest difficulty is in wire bonding process. 41
100 to 1,000 dies can be obtained from one wafer, and 20 to 200 wire bonds per die are necessary. That is, about 20,000 wire bonds are required for one wafer. This is a challenging task. 42
The number of workers in the back-end process is about three times that of the front-end (wafer manufacturing) process. This is the challenge for automation 43
44
In 1960 s, assembling process depended on the manual work of young girls. They were called "transistor girls, and were precious as "golden eggs". It was an era when TVs and radios made by semiconductors were contributing greatly to Japan's exports. As an aside, "transistor girl" also meant a "small and cute girl", and they were very popular, too. Smiles and laughter spread in the hall. 45
Automation has evolved in the following steps. Manual semi-automation full-automation integrated system 46
Outline of integrated system is shown from supply of chips to bonding, molding, and lead-frame cutting. 47
48
This figure shows that productivity improves by orders of magnitude by the shift from manual to semi-automation, to full-automation, and to integrated assembling system. 49
In manual work, the bonding speed is about 1 second per wire, but for full-automation it is 0.2 sec / wire, and further improvement is expected in the integrated assembling system. 50
One of the major effects of automation is improvement in bonding accuracy. Positioning accuracy in manual work is 60μ in 3σ, but it is improved to 25μ with full-automation. 51
52
53
As the feature size becomes finer, the lithography tool changes as follows. Contact method 1 to 1 projection 10 to 1 reduced projection X-ray or EB The biggest problem is a decrease in throughput, and improvement towards the lower right of the figure is needed. From today's perspective (as of 2018), it is a clear mistake to have raised X-rays as a future tool. 54
Regarding memory, "increase in the number of bits" and "increase in volume" occur in parallel. For the gate array, "multi-product, small volume manufacturing" and "short turnaround time" are necessary. Automation must respond to these directions. 55
The trend of the lithographic tools has been "decrease in throughput and increase in equipment cost. The improvement direction must be in the direction of the arrow shown in the figure. 56
Testing throughput decreases due to an increase in test time, pin count, integration density, frequency, etc., and investment cost increases. 57
58
This shows an image of a fully automated factory from order receiving to shipment. The entire process is connected by OA in order processing, by DA in design, and by PA in manufacturing. If you input the wafer, tested finished products are obtained as output. It might have been taken as a dreamlike story at the time, but it is close to the recent image of Industry 4.0, as of 2018. 59
What should people do when factory automation progresses? After instructing the machine, they monitor, support and teach this. Human role is to plan the next thing by thinking by himself. 60
This is a future image of factory automation in a caricature style, and all the work is done by robots. A sales robot, at right hand side, is carrying a box of 64M DRAM, 1,000 times of more density than 64K DRAM at the time, with a lot of sweat not to miss the delivery time to the customer. It is very human-like and invited a burst of laughter. My speech was finished in a big applause from the audience. 61