The semiconductor industry is just now recovering from its worst economic downturn in its history. Yet, companies will scarcely have a chance to catch their breaths before facing a new "crisis of complexity" at 90 nm. This crisis could be a far more daunting
challenge, as it spans process technologies, design tools, and economics. Once again there is a threat to turn ASICs into a "rich man's game," which may represent a major inflection point for innovation in the semiconductor industry as a whole.
For the past 20 years, new IC geometries have steadily appeared every two to three years, delivering the "smaller, faster, and cheaper" benefits that feed the engines of innovation. However, technology challenges at 90 nm are more radical than those of previous geometries. New problems have been created, involving increased crosstalk, high power leakage, and electromigration effects. Many of these problems are multidimensional and span the traditional boundaries between EDA tools, forcing logical and physical domains to be integrated to guarantee timing closure and functionality. At the same time, the economics of ASICs have undergone fundamental changes, as pricing pressures relentlessly drive ever greater levels of functional integration, forcing development costs to skyrocket.
Consider a comparison between 0.25 µm, the leader in design starts in 2000, and 90 nm. Mask costs have jumped from about $100,000 for 0.25 µm to over $1,000,000 for 90 nm. Worse yet, the number of reported re-spins for 130 nm was over 50%, a figure likely to be exceeded at 90 nm. But mask costs are only the tip of the iceberg. The average design of about 5 million gates took about five man-years at 0.25 µm. It is estimated that the designs at 90 nm will average 80 million gates and take over 200 man-years. Overall development costs have grown from about $1.2 million for 0.25 µm to estimates of over $50 million for the average 90-nm design. Software is predicted to exceed 50% of 90-nm costs. As a result, the average size of a design team is growing dramatically, as is the variety of skill sets needed to deliver working deep-submicron silicon. Designers now work alongside DFT, DFM, software, and package engineers, even in the earliest stages of design feasibility.
Moreover, contrast the present day with just 30 years ago, when a few very large companies and IDMs held a virtual monopoly on custom chip development. Only these titans had the economies of scale to provide the required resources: process technologies, design tools, and personnel with critical expertise and know-how. Over the past 30 years, this monopoly has been disrupted by the rise of the EDA industry, pure-play foundries, and the emergence of a global electronics supply chain. The widespread ability to develop custom chips has sparked impressive new levels of innovation, creating countless new products and companies.
Now the convergence of cost and complexity challenges at 90 nm threatens the very engines of innovation. Small- and medium-size fabless companies depend on custom silicon to differentiate their offerings. But if the above trends continue unabated, experts predict that design starts could dwindle to a few hundred, with hundreds of companies possibly disappearing. How many market segments are large enough to support $50 million product development costs?
To counter the complexity crisis, designers are rapidly exploring various alternatives, including low-cost FPGAs, structured arrays with "master slices" of embedded IP, multichip modules, and multiproject wafers. Many companies are also refocusing on their core competencies and partnering or outsourcing to manage design and manufacturing activities. At present, no solutions have emerged as the clear winners in the industry. But clearly, the companies that survive and prosper will be those that adapt most quickly to new solutions and business models as they emerge.
As we transition from the recent economic crisis to the impending 90-nm crisis, we can console ourselves with Farquhar's centuries-old quote, "necessity, the mother of invention." After all, hasn't that always driven new cycles of innovation?
Reprints
