PHILIPS Model 9
New MM9 Extraction routine in UTMOST III
(part 1)
Introduction
In collaboration with STMicroelectronics Central R&D at Crolles (France), a new routine has been developed in UTMOST III to provide a complete solution for MOS Philips Model 9 parameter extraction. This methodology[1] is based on the local optimization method; we can determine a limited set of 18 parameters (so called miniset) to describe the electrical behavior of each device, considering it as the reference device. This miniset includes all the electrical parameters for an individual device[1]. Specific local optimization strategies are described in this article to obtain good minisets. From these minisets, we can extract scaling parameters, using simple linear regressions. We will obtain a new complete set of parameters (so called maxisets). All temperature dependant parameters can also be extracted.
Measures
The MM9 extraction algorithm requires measurements from a Long and Large device, Short devices and Narrow devices (asBSIM3_MG routine). For each device a total of six sets of IV measurements are requested.
Set 1: IDS versus VGS at different VBS and at low fixed VDS.
Set 2: IDS versus VDS at different VGS and at low fixed VBS.
Set 3: IDS versus VGS at different VBS and at high fixed VDS.
Set 4: IDS versus VDS at different VGS and at high fixed VBS.
Set 5: IDS versus VGS at different VDS and at VBS=0V.
Set 6: ISUB versus VGS at different VDS and at VBS=0V.
If these MM9 measurements are stored in a log file and this log file is activated in UTMOST III, then the ALL_DC, ALL_ISUB and AL_IDVGD (ID/VGVD for several devices) can access these measure.
Figure 1: MM9 Measurement Setup Screen
The twenty measurement variables are used and described hereafter.
1.  VGS_start_vg  Starting value for the gate sweep range (ID/VG curves) 
2.  VGS_stop_vg  Stop value for the gate voltage sweep range (ID/VG curves) 
3.  VDS_low_vg  Low fixed VDS bias for the ID/VGVB linear characteristic 
4.  VDS_high_vg  High fixed VDS bias for the ID/VGVB saturation characteristic 
5.  VDS_start_vd  Starting value for the drain voltage sweep range (ID/VD curves) 
6.  VDS_stop_vd  Stop value for the drain voltage sweep range (ID/VD curves) 
7.  VGS_strt1_vd  Calculated starting value for VGS steps for ID/VDVG curve (VBS=0V) 
8.  VGS_strt2_vd  Calculated starting value for VGS steps for ID/VDVG curve (high VBS) 
9.  VGS_strt_off  Offset voltage used to calculate VGS_strt1_vd and VGS_strt2_vd 
10.  VGS_stop_vd  Stop value for VGS steps (ID/VD curves) 
11.  V_source  Constant source voltage 
12.  VBB Maximum  VBS voltage for ID/VG VB curves and high VBS for ID/VDVG curve 
13.  compl_smu(A)  Current SMU's compliance 
14.  points  Number of sweep data points for each characteristics 
15.  VDS_start_gd  Starting value for the VDS steps (ID/VGVD curve at VBS=0V) 
16.  VDS_step_gd  Step value for the VDS steps (ID/VGVD curve at VBS=0V) 
17.  VdstartIsub  Starting value for the VDS steps at VBS=0V (ISUB/VGVD curve) 
18.  wait  Wait time in microseconds, between measurements 
19.  #_of_vgsteps  Number of VGS steps for ID/VD VG curves 
20.  #_of_vbsteps  Number of VBS steps for ID/VG VB curves and for ID/VGVD curve 
Notes:
1. Variables #7 and #8 are calculated as follows:
VGS_strt1_vd = VTH(VBS=0V) + VGS_strt_off, where VTH(VBS=0V) is the extracted threshold voltage for
ID/VGS at VDS low and VBS=0V.
VGS_strt2_vd = VTH(VBS=VBB) + VGS_strt_off, where VTH(VBS=VBB) is the extracted threshold voltage
for ID/VGS at VDS low and BS=VBB.2. The number of VDS step for ISUB/VGS curve is constant equal to 3. The stop value for the VDS steps is VDS_high_vg.
3. The maximum number of points is 201 (variable #14).
4. The maximum value for the variable #19 and #20 is 7.
Miniset Extraction Routine
Miniset Definition
The first step of the MM9 extraction methodology is to extract a miniset of parameters for each device, part of the device strategy selection. For this extraction which will be based on various local optimization strategies, we need to adjust 18 model parameters which will be sufficient to describe the electrical behavior of each device, considered as the reference transistor. These 18 parameters are listed hereafter: VTOR, BETSQ, THE1R, THE2R, KOR, KR, VSBXR, MOR, GAMOOR, ZET1R, VSBTR, VPR, ALPR, GAM1R, THE3R, A1R, A2R, and A3R. All the scaling parameters (SLxx, SWxx, and STxx) must be set to 0.
Figure 2: The six measurement characterization for MM9 Routine.
Clipping Note
In order to obtain accurate minisets, the MOS level 9 model has been slightly modified. The first modification concerns the parameter clipping. We have introduced a new parameter NOCLIP. If set to 1, clipping on THE3R and GAM1R is removed. Negative value for THE3R can be found for long and wide devices. This NOCLIP has been also introduced for GAM1R, to allow negative values during the optimization; but this almost never appears. The user may take care using NOCLIP=1. If THE3R is too negative, solver may not converge.
The second modification concerns the deactivation of the weak and moderate inversion modeling by using ZET1R value, weak inversion factor, as a switch. When ZET1R>5, there is no more subthreshold current. But, the linear region modeling remains the same. A better optimization of the firstorder parameters can be performed.
These two modifications are useful to get an accurate miniset using the local optimization strategies.
User Initial Model Parameter Values
Before launching the extraction procedure, it is important to define the initial value for the model parameters to be optimized. KR and VSBXR model parameters describe the high back bias body effect. These two parameters may not be extracted if only one body effect appears in our technology. For that matter, we can use the ID/VGVB or LGAMMA routine. Using the "Fit" option, and displaying the:
If we do not obtain a straight line, we have to use KR and VSBXR to describe the two body effects. Tables 1, 2 and 3 summarize the initial values of the miniset parameter.
Parameters  1 Body Effect 
2 Body Effect 
KOR  0.6 
0.6 
KR  Not Important 
0.3 
VSBXR  100 
1 
ETAMR; not optimized  1 
2 
ETAGAMR; not optimized  1 
2 
Table 1. Initial values for miniset #1.
In Table 2, these values can be adapted if one strategy does not give good results. This may happen if initial values are really too far from values to obtain. Minimum and maximum values are also important for local optimization strategies. The table below gives you an example of possible minimum and maximum values.
Parameters 
Initial Value 
Parameters 
Initial Value 
Parameters 
Initial Value  
VTOR  0.7  VPR  1.5  PHIBR  0.65  
BETSQ  1.E4  ALPR  0.01  LAP  0  
THE1R  0.1  THE3R  0.01  WOT  0  
THE2R  0.1  VSBTR  100  LER  Mask Length  
MOR  0.5  A1R  3  WER  Mask Width  
GAMOOR  5.E4  A2R  20  LVAR  0  
ZET1R  2  A3R  1  WVAR  0  
GAM1R  0.01  ETADSR  0.6  TR  Temperature 
Table 2. Table 1. Initial values for miniset #2.
Parameters 
Minimum 
Maximum 
Parameters 
Minimum 
Maximum 

VTOR  2 
2 
ZET1R  0 
5 

BETSQ  1.E6 
1.E3 
VSBTR  0 
100 

THE1R  1.E3 
2 
VPR  0.05 
100 

THE2R  1.E8 
0.5 
QLPR  1.E7 
0.1 

KOR  0.01 
1.5 
GAM1R  1.E5 
0.5 

KR  1 
0 
THE3R  0.05 
1 

VSBXR  0 
100 
A1R  0 
30 

MOR  0 
2 
A2R  10 
100 

GAMOOR  1.E9 
0.1 
A3R  0.1 
10 
Table 3. Table 1. Initial values for miniset #3.
All the scaling parameters (whose names begin with SL, SW or ST) must be set to 0. TR will be automatically set to the measurement temperature during the miniset optimizations.
All these initial values must be copied in the User column of the Parameter screen.
Optimizer Setup
The recommended optimizer setup is illustrated in the figure below.
Figure 3. Optimizer Setup recommended.
Local Optimization Strategies
Nine local optimization strategies are proposed. The name given to the strategies is not important, and can be changed without any problem. The strategy number is important. Utmost III will choose by his own which strategy to apply, depending on the type of transistor we are working on. The "Geometry Selected Screen" we can define in the "Target Selection Screen" is not important, as we apply the local optimization only on the device displayed on the "Graphics Screen". MM9 Routine will recognize the type of device we have selected in the Strategy Screen, and then order them to execute the minisets optimization. First we work on the Long&Large device, then on the short devices, and finally on the narrow devices.
We work on ID/VGVB at low VDS for several VBS.
Strategy #1: Linear region for one body effect factor: idvg_mm9_min.
Figure 4. idvg_mm9_min Local Optimization Definition.
Strategy #2: Linear region for two body effect factors Î body_effect_mm9_min.
Figure 5. body_effect_mm9_min Local Optimization Definition.
The optimized parameters are: VTOR, BETSQ, THE1R, KOR, THE2R, KR, VSBXR. During the optimization, ZET1R is automatically set to 100, which will disable the weak and moderate current modeling.
Strategy #3: Weak and moderate inversions: subth_#vds_mm9_min.
Figure 6. subth_#vds_mm9_min Local Optimization Definition.
The optimized parameters are: MOR, GAMOOR, ZET1R. Before executing this optimization, ZET1R is automatically set to a value lower than 5 to enable weak and moderate current modeling. As MOR and ZET1R are strongly correlated, you may adjust this strategy if the result obtained is not good enough.
We work now on the derivative of ID/VDVG at VBS=0V for different.
Strategy #4: gds optimization for long&large device: gd_LongLarge_mm9_min.
Figure 7. gd_LongLarge_mm9_min Local Optimization Definition.
The optimized parameters are: VPR, ALPR, and eventually GAMOOR. For the long&large device, VPR is optimized. LER is automatically set to the length value of the transistor displayed. Like this VPR is identical to Vp. The default GAM1R is 0. GAMOOR has been normally already extracted from the subthreshold region, but may need to be optimized again in that region if more accuracy is needed for low VGS.
Strategy #5: gds optimization for short channel devices: gd_short_mm9_min.
Figure 8. gd_short_mm9_min Local Optimization Definition.
The optimized parameters are: ALPR, GAM1R and eventually GAMOOR. VPR is calculated using the following formula: As for the long&large device, GAMOOR may need a new optimization especially for low VGS. LER is automatically set to the length of the device displayed.
Strategy #6: gds optimization for narrow channel devices : gd_narrow_mm9_min.
Figure 9. dg_narrrow_mm9_min Local Optimization Definition.
The optimized parameters are: ALPR, GAM1R, and eventually VPR. VPR and LER are calculated as for the short devices. Although there is no scaling rule for Vp, VPR may need new optimization for narrow devices.
We work now on ID/VDVG at VBS=0V and for different VGS.
Strategy #7: Saturation current: IDvsVD_mm9_min.
Figure 10. IdvsVD_mm9_min Local Optimization Definition.
The optimized parameter is: THE3R. Take care that minimum value for THE3R may not be lower than 0.005.
We work on ID/VGVB for different VBS at low and high VDS.
(continued in Volume 10, Number 1, January 1999)