Electronic Design

Chip Design, Like All Of Life, Is About Tradeoffs

When I was growing up in India, the game of cricket steadily grew in popularity. Although cricket was introduced to India by foreigners, it caught on with the locals over the past few decades and now they began to beat the people that first introduced it. Naturally, I fell in love with the game and wanted to play. As I started to show some talent in playing the game, my father had a serious talk with me. He taught me that life is truly about tradeoffs. You need to analyze what those tradeoffs are and how they impact your life. So, I analyzed. And here I am in Silicon Valley enjoying a very good career that is "yielding" very well for me.

When things are simple, you can analyze based on rules. When I was a boy, the rule in my house was to get at least an A- and then I could do whatever pleased me. Early in my school years, it was easy to follow that rule and still keep playing. As high school grew competitive, I could not follow the rule. I had been violating the rule, but I didn’t know what I was giving up or what I was gaining for doing that. When my father explained to me what the tradeoff was, it was clear that I couldn’t break the rule anymore.

It is crucial to perform tradeoff analysis. It is not enough to get a go/no-go answer. If the answer is "no go," what can I do about it? And if I decide to live with that answer, what am I gaining? In other words, if I decide to comply to get a "go" answer, what do I have to give up for it?

Let me fast-forward to a few years after my cricket days. By this time, I was busy doing ECL designs at Fujitsu Microelectronics. Then, ECL was used mainly to achieve performance that CMOS could not offer, as speeds of 1 GHz or higher were only possible with ECL. But there were many manufacturing challenges. Our manufacturing experts would characterize earlier chips and create rules that described failing patterns. We would create a floorplan of the logic first, estimate the routing, and analyze the routing patterns for failure during manufacturing. Then, if the failures proved prohibitive, we would change the floorplan and routing to avoid the patterns that were most associated with failures. Once again, I found myself weighing tradeoffs based on rules. Those ECL designs, by the way, were between 800 to 2000 gates total!

Again, when things are simple, you can do tradeoff analysis based on rules. Just imagine doing that same ECL design, but instead of 1000 gates, now you're designing 10 million gates or more on a die of comparative size. Another complicating factor is the number of metal layers. Those earlier ECL designs used two levels of metal; now we are designing with nine or 10 levels of metal.

The beauty of today's semiconductor technology is that you can place 10 million gates on a 17-mm-square die. But, if you get only one good die per wafer, that is of no consequence. Now, there are many design-related and process-related manufacturing issues. What designers need is a way to model these issues and apply that model to their design to analyze their tradeoffs. But designers' lives are already complex. They're juggling area, timing, and power tradeoffs as well as short design cycles. It's hard for them to change what is working for them.

Today’s designers demand evolutionary methodology with revolutionary results. They have been analyzing for area and timing issues to enhance their designs for many years. Then they added power as another parameter a few years ago. Now, if with very little added effort, they could manage the tradeoffs among area, timing, power, and yield, along with some guidelines on how to enhance their designs and ‘what-if’ analysis, they would embrace this new methodology very quickly.

The child in every engineer is reluctant to follow any rules, especially when their life is already so complicated. They don’t want (or need) another thousand pages of rules and a rule-based system that doesn’t give them any practical tradeoffs. They need an elegant solution that evolves their methodologies to provide quantifiable tradeoffs with respect to yield.

And when they tapeout their complex chips with acceptable yields, they are free to play cricket or any other game they like. Now, that is a tradeoff I would love to work for!

Nitin Deo can be reached at [email protected].

Hide comments

Comments

  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
Publish