analog design flow...

[luancao]

I would like to learn analog design. Could you give me a description on the analog design flow ?

Thx

 

[ocarnu]

1 system level (Matlab/Simulink, VerilogA, C, etc).
2 schematic, some hand calculations, simulation, corners, temperature;
3 layout;
4 extract + simukate the extracted (corners, so on);
5a it works, you are happy;
5b it doesn't, see why, back to step 2.

 

[luancao]

ocarnu,

Thx,

I'd like to know what "cornner" means?

 

[kienhuy]

Corners are the extreme conditions for which the design must still meet the design specifications. For example, you may design a circuit with a power supply of 2.5 V, but the circuit should still work for values a little bit above and a little bit below.

 

[holddreams]

tt corner,
ff corner,
ss corner,
sf corner,
fs corner,
......

t means typical,
f means fast,
s means slow.

 

[ocarnu]

The process has some variation. E.g. if Vth nominal is 0.5V (so called "typical value"), actually when you fabricate it might be anywhere between 0.45 and 0.55 with some distrbution. There are lots of parameters and each of them has some variation or uncertainty. Corners take the extremes, like most unfavorable case and most favorable (which is which depends on design). If you don't simulate the corners you'll have the surprise that only 50% of the chips work. That means double the price for the chip (more actually, you have to test them to find the good ones). That means you go bankrupt .
In some designs corners are not enough. You have to take more samples from the parameters space, with a monte carlo analysis. How to take these parameters to cover efficiently the whole space, without missing the important spot, it's a science in itself.

 

[luancao]

thx, I have been a digital guy for many years. How can I convert digital to analog?

What books or courses shall I learn?

 

[mince]

thx, I have been a digital guy for many years. How can I convert digital to analog?

What books or courses shall I learn?

Analysis and Design of Analog Integrated Circuits by Gray, Hurst

 

[selvaraja]

Analog Integrated Circuit by Behzad Rahzavi

 

[vale]

1 system level (Matlab/Simulink, VerilogA, C, etc).
2 schematic, some hand calculations, simulation, corners, temperature;
3 layout;
4 extract + simukate the extracted (corners, so on);
5a it works, you are happy;
5b it doesn't, see why, back to step 2.

I'm unhappy now because simulation result of netlist extracted from layout is quite different from the design!

 

[luancao]

Then what shall you do next?

And BTW, what's the post-layout netlist?

 

[ferrite]

i think post-layout netlist is the extracted netlist
from your layout. when you do the first hand netlist simulation, you generally ignore some hidden nodes, and parasitics. I think in the post-layout netlists, those hidden stuff get counted to a certain level such that your simulation could make more sense

 

[tlihu]

I'm unhappy now because simulation result of netlist extracted from layout is quite different from the design!

So, this is where experience counts for analog design.

 

[flysnows]

thx, I have been a digital guy for many years. How can I convert digital to analog?

What books or courses shall I learn?

cmos integreted circuit design by P.E.Allen

 

[holddreams]

The book of ken martin is also helpful.

 

[Areky_qin]

but how to simulate the system level circuit?

 

[edk]

in cadence....spectre...u can do that...but there transistor count is an importatnt factor...if much transistor are there...it will take many hours...
edk

 

[alenhan]

maybe you also need to learn Hspice

 

[suria3]

The process has some variation. E.g. if Vth nominal is 0.5V (so called "typical value"), actually when you fabricate it might be anywhere between 0.45 and 0.55 with some distrbution. There are lots of parameters and each of them has some variation or uncertainty. Corners take the extremes, like most unfavorable case and most favorable (which is which depends on design). If you don't simulate the corners you'll have the surprise that only 50% of the chips work. That means double the price for the chip (more actually, you have to test them to find the good ones). That means you go bankrupt .
In some designs corners are not enough. You have to take more samples from the parameters space, with a monte carlo analysis. How to take these parameters to cover efficiently the whole space, without missing the important spot, it's a science in itself.

Hi Ocarno,

I read your reply. That means that when we design a circuit, we will run the Typical, Worst and Best case simulation over temperature. And a part of that, which simulation is the best to say that can guarantee that our design chip will work after fabrication, is that Monte Carlo or Corner analysis? Then is Typ, Worst & Best analysis alone not comparable with monte carlo or corner. Which will you recommend to guarantee our chip work. Like when i run monte carlo (process & mismatch) at typical case for 50 iteration, i found out about 5 nodes going out of specifation, so at what percentage we say the monte carlo analysis pass our design. Hope to get a good feedback from you.

Thanks,
Suria

 

[beckchm]

it depends on the configuration and complexity