Professional Documents
Culture Documents
Super Buffer
Super Buffer
EE-166
Super Buffer
We know that to drive large loads we increase the widths of our transistors. How do we drive these larger transistors?
Scale a chain of inverters.
Cd (ln 1) = Cg
EE-166
Super Buffer
How do we know how many to use in the chain and how to we scale up? If the Drain Capacitance is close to zero, the scaling factor reduces to e=2.718. What do we do when it is not?
Solve numerically for .
Cd (ln 1) = Cg
EE-166 3
Super Buffer
Rearrange the design equation.
A=
Cd + Cg
Example
You have a NAND3 that needs to drive a 1000 identical NAND3s with a minimum propagation delay.
The output of the buffer should not logically invert the out put of the NAND3. Assume that WP=1.5m and WN=1.8m and that Ln and Lp are minimum sized and that the propagation delays for the worst case transitions are symmetric for the NAND3. Assume that this will give you symmetric rise and fall times.
EE-166 6
EE-166
EE-166
EE-166
EE-166
10
EE-166
11
Calculate WN and WP
Remember we scaled WN by 3 to take into account Leff=3LN. We divide by 3 to get a regular inverter. Ignore the 0.
EE-166
13
Calculate the delay for each stage and the total delay.
EE-166
14
EE-166
15
Use calculated capacitance. 1000 NAND3s would take a long time to simulate.
EE-166
16
Schematic
EE-166
17
Results
The propagation delays were ~1ns with 4% difference.
This was 14% slower than we predicted. Should we adjust the A constant by 1.14?
If we do then the error between our hand calculations and our spice results drops to 2%.
The rise and fall times were ~600ps with 10% difference. The power consumption was ~60mW.
EE-166 18
Layout
This is tricky because we are going from something that is small to something that is large.
Keep the cell height the same as the smallest inverter and use multipliers
This will cause the last buffer to have at least 6*6=36 fingers which will lead to a long poly line
INPUT OUTPUT Cell Height
Width changes by n
EE-166
19
Layout Continued
Keep the same cell height but break up the last inverter into manageable chunks chunks
INPUT
OUTPUT
This causes the in and out ports to be on the same side. (Maybe you want this.)
20
OUTPUT INPUT
OUTPUT
OUTPUT
EE-166
22
EE-166
23
EE-166
24
EE-166
26
Extract/LVS
EE-166
27
The final buffer was split into 2 instead of three because it gave an even number of fingers.
EE-166
28
EE-166
29
Final thoughts
Are the poly lines too long?
Antenna rules
Are lines wide enough? What would the power be if we had 40 output pins switching at the same time?
EE-166
30
Results (extracted)
The propagation delays were ~1ns with 3% difference. The rise time was 509ps and the fall time was ~600ps (14% difference.) The power consumption was ~53mW. The total area was 53 by 72 m.
EE-166
31
EE-166
32
VIU
VIU
VID
EE-166
33
VID
VIU
EE-166
34
Alternate Implementation
EE-166
35
Final Schematic
Propagation Delay was .49ns VTRIPHL=.976V 2% Error VTRIPLH=3.13V 4% Error
WNI 8.5 WPI 22.1 WNF WPF 8.78 133.8
EE-166
37
VTRIPLH
Analysis
EE-166
38
VTRIPHL
Analysis
EE-166
39
LAYOUT
EE-166
40
EE-166
41