U2U keynote, Rhines
April 26, 2011, Mentor U2U, Santa Clara,CA—Wally Rhines described the dynamics of the design automation industry at the Mentor users' group meeting. The EDA industry has fairly unique characteristics as a marketplace that seem to defy description and analysis.
Mentor is growing and approaching $1B in total revenues. Some ask if EDA tracks industries like the semiconductor industry. Most analyses suggest some relationship, but the correlation coefficient between semiconductor revenues and EDA is 0.15. The semiconductor business seems to create its own cycles, independent of the state of the world economy or of any single country.
The semiconductor industry moves from over to under capacity and has many pricing fluctuations. One statistic of worth is silicon area, which is getting back to the normal growth curve after dropping in Q1 of '09. Unit shipments are also on a growing trend line which has been relatively constant except for '01 when the industry collapsed in the .com bust, and '09 due to an inventory correction. Revenues for the semiconductor industry are looking to exceed $300 B and represents a 32 percent gain in '10 over '09. indications are that the industry will grow more this year.
The disasters in Japan will affect the supply chains in the world, but represents only 22 percent of the world capacity, or 2,714,000 wafers per month. The projected increases of over 1 percent is more likely to end up as a decrease of 2 percent with revenues dropping from $4.3 B to $4.22 B. There will be some effect on world GDP, but with the fabs coming back on-line, the drop will be fairly small. This will be similar to the situation when Sumitomo chemicals has a fire that affected the world's supply of encapsulation epoxy.
The Japanese are good at optimization and at that time quickly moved to develop substitute supplies. In addition, the capacity utilization was well below capacity, so effects were minimal. Now, analysts are still forecasting growth for the Japanese electronics companies for this year. The shortages will be short term only.
These events demonstrate the benefits of specialization. In the "80's, Japanese companies all made everything. Now, companies are focusing on a few product lines. The problem is that specialization creates worldwide dependencies. Some examples are: specialized oxides for paint will make black cars hard to get this year, Lithium ion polymers for batteries and PCB substrates will be in short supply for a while, and over 50 percent of automobile air flow metering components are produced in Japan.
Only two companies in Japan are responsible for over one third of all LCDs. 90 percent of products in the critical flow path are from a sole supplier in Japan. Between '90 and '10, many companies installed a just-in-time production flow with sole sourcing for many parts. This change reduced work in progress and inventory at the risk of a single component shortage shutting down the whole line.
The earthquake is creating an industrial backlash to single sourcing and making manufacturers to consider going back to multiple sources. The value of multiple sources is geographic diversity as well as more cushioning in the supply chain.
The problem with dual or multiple sources is that costs increase for everyone, while R&D efficiency and the rate of quality improvement decrease. As an example, Alcoa was the sole supplier of aluminum in '48. The government brought in Kaiser and Reynolds as alternate sources due the war needs and reduced its risk with that intervention. The result was that indeed, R&D costs went up and the prices stayed flat. Before the war effort, Alcoa was innovating and reducing prices to try to compete with the alternative materials like steel. Their new competitors provided a floor for prices and reduced the need to innovate.
EDA is in need of greater concentration. There are 69 market segments in EDA that bring in over $1M per year. There are over 1,000 companies serving those markets and an additional 10,000 providing services. The big companies dominate the biggest segments and the little companies control the little markets. The top supplier in each category averages 66 percent of the market and all the leaders have at least a 40 percent share. Across the EDA space, the number one positions are fairly evenly distributed, but the biggest three vendors control about 73 percent of the total EDA market.
But the EDA industry continues to grow. VC's have cut down the total dollars which peaked at $500M in '00, but startups continue to create churn in the industry. About 50 new companies start every year, with about 10 to 15 getting acquired within a few years of starting. About 20 companies fail each year still leaving 15 companies remaining. The number of companies in EDA continues to grow.
This growth causes a dispersion of R&D efforts and resources, even though the EDA companies collectively have enabled an increase of 4 orders of magnitude in transistors per engineer per year over the past 25 years. EDA has remained about 2 percent of semiconductor revenues over the same time period. Meanwhile, the cost per transistor has dropped about 35 percent per year and the cost of EDA tools per transistor has also dropped.
The problem is that the EDA market remains fairly flat with about 2 percent growth while the number of people in EDA is growing at over 5 percent per year. Mentor's market share has grown from 15 percent in '98 to 24 percent in '10 and the growth has come from solving new problems. New methodologies drive increases in EDA revenues.
The next challenges for the design industry include: high-level design with sub-areas of design exploration and performance and power estimation being key. High-level design must address the architectural level of a system through transaction-level simulations on virtual prototypes. These design components then go to high-level synthesis and verification.
Functional verification and intelligent test benches are necessary to increase the coverage by 10X as well as make the verification processes finish faster. This is a linearly scalable problem that can benefit from multi-processors. Another area of interest is hardware acceleration, with the driver being the growing size and complexity of the latest chips which is overwhelming the verification tools.
A change to a transaction-based test-bench can resulting a 50 to 500 X improvement in hardware-software co-emulation. The simplification of the emulation process through virtualization enables a move of emulation to the cloud, removing the hardware cost as an impediment to greater adoption of emulation.
Physical tools must integrate the place and route with verification and the design rule sets to improve throughput. A programmable electronic rules check is a critical component and physical device parameter extraction must change to a full 3-D mesh for the necessary accuracy needed at the next process nodes.
Embedded software is becoming another elephant. Out of a $100 M total design cost, the hardware costs are remaining fairly flat while the software costs are doubling. Now a large project has twice as many programmers as hardware developers and the costs for software and integration at the systems level is twice the hardware development costs.
To fix this problem, embedded software has to move to more automation and reuse, just like the hardware has done already. In addition, open standards for software will open up more areas for automation.
Specialization has benefited all semiconductors and EDA specialization will drive more designer productivity.