It was a quiet week in EDA until yesterday when a headline on the EE Times website caught my eye. It stated: "Semicon panel: Design-for-manufacturability no longer a luxury".
In 1968 at the TRW Microelectronics Laboratory we were designing ICs with features so big by today standards, that if you just held the die at the correct angle with respect to the sun, you could see most of the devices in it. And yet, it never occurred to any of us that we should not consider manufacturability during the design. Of course we were a true IDM, so we controlled every aspect of the flow, including all of the CAD tools.
Then in 1970, working on one of the first commercial microprocessors for a Los Angeles company long gone out of business, we used both Texas Instruments and AMI as our foundries. During those designs, we spent many hours talking visiting with them to understand what could be manufactured with acceptable yields, what could be tested, and what could not.
Now in the twenty first century, it takes a panel of industry leaders to tell us that it is wise to design stuff that can be manufactured. It seems that design methods have gone backwards, not forward with time. I agree that it may sound manly to cram as many transistors as possible in the smallest possible silicon area in the futile attempt to discover how many angels can fit on the head of a pin. But it is much manlier to try to understand the manufacturing process so that we can design stuff that will actually have an acceptable yield while obtaining the best possible performance metrics.
One of the few professional satisfactions an engineer can have is to be able to totally blame the sales and marketing guys if a product is not profitable while pointing with pride to the performance/yield records of the product. Of course things have gotten more difficult. There is no question that engineers today are confronted with a different set of physics problems than thirty-five years ago: yet the goal has remained the same. The difference is that then we felt the responsibility of learning about the manufacturing and testing problems the IC would face once we were "done". Unfortunately today the industry highlights specialization at the detriment of knowledge of the entire process.
Just because something is hard, it is no excuse to ignore it. The reason foundry need to set design rules is that designers consistently try to test the limits of physical laws in the hope that, in their case, mother nature will make an exception.
So, whether we called Design For Manufacturing, or Design For Manufacturability, it is really not important. It should be Manufacturing Aware Design. Put the important name in front to give it emphasis. So the acronym is MAD and not DFM: it is not so bad, the complexity of the task does tend to drive people mad. And it is mad not to do it. So, let's just call it what it is, shall we?