Extracting BSIM3 Model Parameters Using UTMOST

 

Introduction

As was detailed in a previous issue of The Simulation Standard, the BSIM3 Version 2.0 model has been incorporated into UTMOST. Up to now, users could extract BSIM3 model parameters using either user-defined global or user-defined local optimization strategies. UTMOST now includes a dedicated in-built BSIM3 parameter extraction algorithm which is based on the extraction methodology proposed by Berkeley. The UTMOST BSIM3 extraction routine can perform both the measurements and the extraction sequence necessary for the extraction of an accurate scalable BSIM3 model. This article will describe the BSIM3 measurement and extraction routine. As an example, the results of using this routine to extract n-channel and p-channel scalable models to devices with effective channel lengths down to 0.6µm will be shown.

 

BSIM3 Measurements

The BSIM3 extraction algorithm requires measurements from (a) a device with a large drawn length and a large drawn width (a Large device), (b) devices with a large drawn length and different drawn widths (W-array devices), and (c) devices with a large drawn width and a range of drawn lengths (L-array devices). The number of devices in the W-array or L-array is user-definable and UTMOST will accept a minimum or one device in each array. For a truly accurate scalable model it is recommended that a minimum of two devices exist in the L-array.

 

A total of four sets of I-V measurements are recorded by the BSIM3 routine for each device. These measurement sets are detailed below.

 

  • Set 1: IDS VGS measurements where the gate voltage is swept between defined minimum and maximum values. The source voltage is grounded and the drain voltage is set to a low value (i.e. 0.1V). The bulk voltage is stepped, in six steps, between 0V and a defined maximum.

 

  • Set 2: IDS-VDS measurements where the drain voltage is swept between defined minimum and maximum values. The source voltage is grounded and the bulk voltage is set to a low value (i.e. 0V). The gate voltage is stepped, in five steps, between defined minimum and maximum values. The minimum gate voltage is determined using the threshold voltage calculated from the data measured in measurement Set 1.

 

  • Set 3: IDS-VGS measurements where the gate voltage is swept between defined minimum and maximum values. The source voltage is grounded and the drain voltage is set to a high value (i.e. 5.0V). The bulk voltage is stepped, in six steps, between 0V and a defined maximum.

 

  • Set 4: I DS-VDS measurements where the drain voltage is swept between defined minimum and maximum values. The source voltage is grounded and the bulk voltage is set to a high value (i.e. -3.0V). The gate voltage is stepped, in five steps, between defined minimum and maximum values. The minimum gate voltage is determined using the threshold voltage calculated from the data measured in measurement Set 1.

 

Most of the voltage settings associated with the BSIM3 measurements are user definable. It is important to note that the UTMOST BSIM3 routine allows the user to decouple the measurement and extraction tasks. This is recommended because probe time is minimized and the user can exercise the maximum amount of flexibility with respect to the extraction of the BSIM3 parameters. The BSIM3 measurements should be stored in an UTMOST log file and used as required. In this case, all of the data measured and used by the BSIM3 routine is also available for use by UTMOST's general purpose ALL_DC routines. Thus, needless re-measurements are avoided in situations where the user wants to perform user-defined regular or local optimization strategies using the BSIM3 data. The ALL_DC routine is also very useful for model validation. A typical plot of the BSIM3 measurement data, as viewed by the ALL_DC routine, is shown in Figure 1.

 

 

Figure 1. Typical BSIM3 measurements for a n-channel device.

 

 

Parameter Extraction

Prior to any parameter extraction the user should specify values for some of the so-called BSIM3 elementary parameters like TOX, XJ, NPEAK, and NSUB. BSIM3 expert parameters can also be specified. The UTMOST BSIM3 extraction algorithm will now be described.

 
Step 1: Use the Large device linear region data from data Set 1, and threshold voltages calculated from this data, to extract VTH0, K1, K2, U0 (optional), UA, UB, and UC.
 
Step 2: Use the W-array linear region data from data Set 1, and threshold voltages calculated from this data, to extract K3, DW (optional), and W0.
 
Step 3: Use the L-array linear region data from data Set 1, and threshold voltages calculated from this data, to extract NLX, DL (optional), DVT0, DVT1, and DVT2.
 
Step 4: Use the Large and L-array subthreshold region data from data Set 1, and subthreshold slopes calculated from this data, to extract VOFF, NFACTOR, and CDSC.
 
Step 5: Use the L-array and W-array linear region data from data Set 1 to extract RDS0 (optional) and RDSW.
 
Step 6: Use the Large and L-array data from data Set 2 to extract VSAT, A0, A1 (if BULKMOD = 2), and A2 (if BULKMOD = 2).
 
Step 7: Use the Large and L-array data from data Set 4 to extract KETA.
 
Step 8: Use the L-array saturation region data from data Set 2, and the output resistances extracted from this data, to extract PCLM, DROUT, PDIBL1, PDIBL2, PSCBE1, and PSCBE2.
 
Step 9: Use the L-array subthreshold region data from data Set 3 to extract ETA0, ETAB, and DSUB.
 
Step 10: After the extraction algorithm is completed UTMOST will optimize any selected BSIM3 parameters to the saturation region output conductance data for the L-array devices. In some cases this will improve the quality of the extracted parameters considerably. If the user does not select any parameters for this parameter refinement stage then this step will be skipped.

 

The extraction of the U0, RDS0, DL, and DW parameters is optional. The user can set them to pre-defined values if required. By default the BSIM3 BULKMOD parameter will be set to 1 for n-channel devices and 2 for p-channel devices. However, the user has the flexibility to override these values and set BULKMOD equal to 1 or 2 for either n-channel or p-channel devices. An example of a BSIM3 linear region fit for an n-channel device is shown in Figure 2 while a BSIM3 saturation region fit for a p-channel device is shown in Figure 3.

 

 

Figure 2. Example of a BSIM3 linear region fit for an n-channel device.

 

Figure 3. Example of a BSIM3 satruation region fit for a p-channel device.

 

 

 

Example

The UTMOST BSIM3 extraction routine was used to measure data and extract a scalable model parameter set for a range of devices which included device geometries (W/L's) of 20/20µm, 3/20µm, 20/0.65 µm, 20/1µm, and 20/2 µm. Both an n-channel extraction and a p-channel extraction were performed. The extractions proved to be quite accurate as can be seen from the measured versus simulated device characteristics in Figure 4 (n-channel) and Figure 5 (p-channel), so global or local optimization was not necessary.

 

 

Figure 4a. Some measured ( ___ ) and simulated ( ---- ) L-array device characteristics for
n-channel MOSFETs after a BSIM3 extraction.

 

Figure 4b. Full set of measured ( ___ ) and simulated ( ---- ) device current characteristics for
a 20/1µm n-channel device after a BSIM3 extraction.

 

Figure 4c. Measured ( ___ ) and simulated ( ---- ) device currents and conductances for
a 20/1µm n-channel device after a BSIM3 extraction.

 

Figure 5a. Some measured ( ___ ) and simulated ( ---- ) L-array device characteristics for
p-channel MOSFETs after a BSIM3 extraction.

 

Figure 5b. Full set of measured ( ___ ) and simulated ( ---- ) device current characteristics for
a 20/1µm p-channel device after a BSIM3 extraction.

 

Figure 5c. Measured ( ___ ) and simulated ( ---- ) device currents and conductances for
a 20/1µm p-channel device after a BSIM3 extraction.

 

 

These plots were generated with the ALL_DC routine which reads data recorded by the BSIM3 extraction routine. The predictions of device current and conductance in all regions of normal device operation are in very good agreement with the actual measurements.

 

It was decided to hold the contact resistance parameter at a set predefined value suitable for the devices under test. The BULKMOD = 1 model was chosen for the n-channel devices while the BULKMOD = 2 model was used for the p-channel devices. In these examples no use was made of the UTMOST BSIM3 extraction facility whereby model parameters can be refined or tuned using optimization after the regular BSIM3 extraction. This facility would have improved the model accuracy even further if it had been used. In addition, user-defined local optimization routines or any of the regular UTMOST optimization features could also have been used to refine the BSIM3 parameters using the data measured by the BSIM3 extraction routine [2,3].

 

Conclusions

UTMOST now has a very powerful in-built BSIM3 measurement and extraction routine. Scalable BSIM3 models can easily be extracted with this routine. Extraction examples for n-channel and p-channel devices from a 0.6 µm CMOS process were used to demonstrate the effectiveness of the extraction algorithm. The BSIM3 extraction routine includes an optional parameter optimization facility which can be used to refine or tune the extracted model parameter set. The data measured by the new BSIM3 routine can also be accessed by the general purpose ALL_DC routine. This means that flexible plotting and optimization features are available for the measured BSIM3 data subsequent to the BSIM3 extraction.

 

References

[1].J.H. Huang, Z.H. Liu, M.C. Jeng, K. Hui, M. Chen, P.K. Ko, and C. Hu
BSIM3 Manual (version 2.0)
University of California, Berkeley, March 1994.

[2]. UTMOST Application Note, Reference No. UT/94/004
Extraction of MOSFET BSIM3 Parameters with UTMOST using Global Optimization Techniques.

[3]. UTMOST Application Note, Reference No. UT/94/008
Extraction of MOSFET BSIM3 Parameters with UTMOST using Local Optimization Techniques.