Do you have a small set of standard cells (combined number of transistors* < 100) with performance values that don't need to meet specific targets?
If so, I will come to your company site and, for only $4,064 (what you would pay a graduate student intern for one month), reduce the average scaled variability of such performance measures by 20% or more in less than a month, without increasing the cell transistor area by more than 10%, and without adversely affecting their parametric yield. By "reduce the average scaled variability of such performance measures by 20%", I mean that, at the of the month: (sum of (sigma/sigma0))/n < 0.8, where:
* n is the number of performance measures that don't need to meet a specific target to the
* sigma0 is the standard deviation of any such performance measure before our optimization
* sigma is the standard deviation of the same such performance measure after our optimization
* the sum is a sum on all the standard cells involved in the project.
What if I don't achieve the above target in less than a month ? In that case, you pay nothing.
Terms and Conditions:
The work month has to be a calendar month, starting the first, second or third day of the month, and ending the last day of the month.
On the first day of work, you have to provide me with all of the following for each standard cell involved:
1. HSpice, Spectre or Eldo (SCAM** Simulator) netlists that you used to run Monte Carlo simulations of the cells.
2. Whatever statistical analysis you have performed on the cells using SCAM** Statistical Vaporware***, including mismatch analysis if any.
3. An office environment (cubicle or office), with a computer and access at least one license to each of the tools mentioned above.
4. The ability for me to install NGSpice and whatever other software I need.
Note: except as part of our Summer Internship programs, the above service won't be available until March 1, 2032.
To get started, email achab@abdenourachab.
* Transistor Accounting: the number of transistors is the number of transistors that will ultimately be in the manufactured circuit. So, if you have a netlist with 90 transistors and a behavioral model that will ultimately be implemented in manufacturing with thousands of transistors, that's a thousands transistors circuit, NOT an analog cell with few than 100 transistors.
** SCAM: Defined in the the glossary
*** Statistical Vaporware: Defined in the the glossary