lastdance said:Emitter degeneration is wasteful as it involves a resistor. To increase the output impedance of a current mirror, cascoding is almost always good enough, unless your voltage range is very wide. Very often, it's not the early effect but the inherent mismatch that kills, hence, the key is to use units and large devices to improve matching.
lastdance said:Inherent mismatch are mismatch due to the process, you dont see it on sim.
In current mirror, you can have geometrical mismatch, for eg., you want W to be 5u, but due to the process tolerance, it could end up to be 4.8u. To improve matching, you draw larger devices. You also have beta and Vth mismatch which are often the dominant mismatch parameters. Therefore, you need to design for larger overdrive to reduce effect of these mismatches. If you could, do common centroid. I normally do not use dummies, they dont seem to help.
What I meant was to draw larger devices in units. It's true that if you break a large device into many small unit pieces, you get the averaging effect and hence better matching. But I dont think you should use minimum size devices when it comes to accuracy.
terryssw said:I think dummy devices is very important in matching, since your external devices will see different boundary condition thus different etching rate compared to your internal devices.
By the ways, how should we divided a devices to have the greatest matching performance? For example I want to match two 1pF capacitor, should we divided it into units of 500fF, 200fF, 100fF, 50fF or even smaller? Do it really the smaller the better??
alchen77 said:terryssw said:I think dummy devices is very important in matching, since your external devices will see different boundary condition thus different etching rate compared to your internal devices.
By the ways, how should we divided a devices to have the greatest matching performance? For example I want to match two 1pF capacitor, should we divided it into units of 500fF, 200fF, 100fF, 50fF or even smaller? Do it really the smaller the better??
I don't think smaller capacitance has better match, the reason we break the capacitance :
1. decrease the well or diffussion resistance.
2. layout can mixed these two capacitances for better match. if you break the capacitance but not mix it. not good.
3. always add dummy patern in the layout boundary.
4. process variation makes small percentage in the big componet.
Vamsi Mocherla said:Definately, breaking a larger device into smaller pieces in layout will surely help. Only constraint is the area. In design, it is always advised to have a wider device because it helps the device to hold good even in corner conditions
Vamsi Mocherla said:Well smaller size is the size where the unit transistor(or any other cell) can give reliable results .
Consider if we need a device of W1/L1 for a current of I1. You should be able to set a device of size W1/L1*1/K which will reliably give a current of I1/K. Hence K would be the number of fingers
terryssw said:Vamsi Mocherla said:Well smaller size is the size where the unit transistor(or any other cell) can give reliable results .
Consider if we need a device of W1/L1 for a current of I1. You should be able to set a device of size W1/L1*1/K which will reliably give a current of I1/K. Hence K would be the number of fingers
Can you discuss more on how reliable should be, or how you can measure if the device give reliable current? Since larger unit devices will always produce more reliable current, but we always preferred smaller unit devices in terms of matching? Is there some explicit limit on the choice of unit devices?
pbs681 said:If we break big resistor into smaller, we will have more contact right. ANd this contact has resistance, therefore will incerase the overall resistance value.... Seems like problem.... any comment
We use cookies and similar technologies for the following purposes:
Do you accept cookies and these technologies?
We use cookies and similar technologies for the following purposes:
Do you accept cookies and these technologies?