|
发表于 2011-1-14 15:26:21
|
显示全部楼层
本帖最后由 fhchen2002 于 2011-1-18 06:38 编辑
回复 4# domino.kk
1. Do not use wide width device. (for instance,
W_min = 0.24 um,
Width with single contact = 0.42 um,
Width with two contacts is 0.89 um.
Choose width = 0.89 um, and no more wider. <- Choose a MOS device, without significant narrow-width effect, but narrow enough.
2. Choose as long channel length as possible.
3. If possible, use pMOS as the unit-current mirror device, instead nMOS.
The reason is the mobility of pMOS is smaller. With the same bias current, the (VGS - VT) can be effectively larger.
Steps 1-3 are enforced to make maximum VGS - VT possible.
4. Determine Large enough VGS - VT.
5. Employ cascode to increase the output impedance of each unit current mirror.
Example:
Single ended output
Full-scale current output: 1023 * 5 nA ~= 5 uA
R_load = 200 KOhm (full-scale: ~= 5 uA * 200 KOhm ~= 1.0 V)
When all current sources are in parallel, you see r_out / 1,023 of the 1023 unit-current sources.
Make sure this number is much greater than 200 KOhm.
This means your unit-current source needs to have r_out >> 200 MOhm
6. Separate your current mirror units from the decoder circuits. Put all your current mirrors together and make 2-D scramble. This is to minimize the impact of process-dependent gradient.
7. Put at least two columns of dummy current mirrors at the array edge (both sides) to minimize microloadings.
8. Depending on DNL and INL design specification, determine binary weigthted current mirror and thermometer coded current mirror.
寫得比實做容易
|
|