You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
After reading this, I have questions about how broadly Full Stack Experimentation can be applied to a number of areas including EDA in semis. Any opinions thoughts or comments would be welcome as this is beyond me. It seems it might be very time consuming, except with computational power increasing along with memory, this might not be a factor. Thanks
The article talks about consumer-oriented software where the user experience is paramount, something that doesn't make sense for EDA software which is engineering software often with very little UI, think command line for many EDA tools (DRC, LVS, DFM, SPICE [except for analyzing waveforms], P&R [except for floorplanning], logic synthesis, RTL coding, C++ coding).
Dan, I know it's an entirely different application, but I was more thinking about the basic structure and layout. Couldn't this be applied to EDA of just about anything? It looks just like a different and more broadly based form of taking information from a much broader base. Could you give some insights in a post to the changes in EDA you see coming in the future as to new structures, approaches and the future application of AI/ML? Thanks
Dan, thanks, it almost sounds like the financial program I have been working on and tuning for years. I have studied Monte Carlo simulations and they are extensively used in financial and gambling programs. Renascence Medallion, the world's most successful financial construct uses state of the art AI/ML with highly advanced math and computational systems and has the best return over a twenty plus year period. I find some new way of looking at my financial model every week. I wish back testing worked, but for me it doesn't. I do all my test real time with real money in large increments for only the real world testing works because of transaction size, frequency, timing and outside considerations. I find paper only tests to be lacking in many respects and haven't had a losing trade in a long time. I do have one set of losing positions on an early experiment that I will be ending as a needed tax adjustment tool. This is all I do for a living and my hobby. I used to read science fiction voraciously and now I live it. Never underestimate new and novel approaches to anything. I did once and now have a 60K tax loss sitting and waiting to remind me of my mistake. I make at least two mistakes everyday, but get enough right to do well, it's just odds and math.
