Wednesday, November 23, 2005

Variations : I hate those

I am a kind of uniform person. I do not like a lot of variations, may those be anything : right from my schedule to the food I eat, or for that matter, the fonts I prefer. I hate variations.

On my last day at IISc, our juniors treated us at the Student's Amenities hall. It was a grand treat, with good food, music and everything. It was a memorable party. One of the "items" on the itinerary was "Guess Who?". The theme was simple, the juniors would enact peculiarities of one of their seniors and we had to identify who he or she was amidst the huge laughters. And guess what, they started off with me!
Now this "guess who" fellow would start his day at sharp 6 O'clock in the morning, go and play badminton for precisely so-and-so time, come back to room, take bath for exactly fifteen minutes, and rush to his lab after a pre-calculated breakfast at the C mess. No matter what day of the week it is. This fellow was hailed "Salimbhai HMT" (I don't know why they had to append a a bhai to so romantic a name as Salim, but later I realised all others' names also sounded rowdy!). And boy, that was me.

I have enjoyed consistency. Whether it be in studies or sports or anything worth while I have been involved in. I remember I never missed my Sanskrit tuition class in standard 10: the tuition class would start exactly at quarter to six in the morning, whether it be raining or snowing outside. And I was proud of that. When I commented on that, my mom was proud : She told me of her achievement and how her mother was proud. So, after all, it runs in the family!

Now why all of a sudden I am writing about this is that I just came back from a talk on "On-chip variations and techniques to reduce the ill effects". It was really a good talk, more so, since my work is involved closely with the implications of on-chip variations.

One fascinating fact from the talk (pardon me if I am not able to explain it well for those who are not "electronic"ally inclined).

Basically, with the transistor sizes getting smaller and smaller by every two years, we are getting more and more of them per square millimeter of silicon area, for less cost. Yet, the problem is that we are not able to produce those identical to each other. Obviously, each one of a billion transistors cannot be expected to behave identically to his other one billion minus one brothers. So we have the problem of variation on the chip, that hits performance and power consumption more than other metrics.

Now here is one immensely interesting cause of variation :
One transistor may contain a few hundred dopant atoms (atoms of Boron or some other metal) mixed with silicon in a transistor. After a few years, this "a few hundreds" number is expected to go down further "to a few tens". In such case if one of the transistor gets one extra atom than its neighbouring transistor, we have considerable variation in the performance. Isn't that amusing.

But it doesn't stop at that.

Even if two transistors have equal number of these impurity atoms, still we may end up in variation in performance. This variation is because of mismatch in relative locations of the impurity atoms in the two transistors!

How's that for expectation of uniformity?

And look at me, I can go out of the usual routine and skip my daily evening cup of coffee if required :)

No comments: