Chapter 3 Four-port RF MOSFET Modeling for Simulation with DBB ( UN65 CMOS
3.2 Improved Body Network Model for Four-port RF MOSFET with DBB
3.2.2 Improved Body Network Model-Equivalent Circuit and Extraction Method
According to the comments made in 3.1.1, the simple body network model proposed for L90709 (Fig. 3.2) cannot be applied to L65003, owing to different layout in the interconnection to p-well body and deep n-well. Due to the fact, a new body network becomes indispensable for 4-port RF MOSFET in L65003. First, the device structure of the 4-port MOSFET designed in L65003 is illustrated in Fig. 3.27 to facilitate the body network model development. Herein, Rbb represents the body resistance associated with p-well, Rdnw is the series resistance going through the deep n-well, and Rbb2 as well as Rbb3 denote p-substrate resistance. Regarding the capacitive components, Cjs and Cjd are well known as the source and drain to body junction capacitances. Cdnw1 and Cdnw2 define the junction capacitances from deep n-well to p-well and p-substrate, respectively. Based on the device structure and RC components allocation defined in Fig. 3.19, an equivalent circuit can be derived as shown in Fig. 3.20. Note that the RC network in the solid-line box is composed of Rbb3 in parallel with the series Rbb2Cdnw2 and the equivalent impedance is defined asZsub. As for the RC network in the dash-line box, it is consisted of Rbb in parallel with the series RbbCdnw1 and the equivalent impedance is defined as Zbb'. The idea underlying the proposed RC network comes from the strong frequency dependence of Re(Z44)=Re(Zsub) as shown in Fig. 3.21, which was measured from port-4 (body) when all the other 3 ports (1 : G, 2 : D, 3 : S) are at open state. The fast fall off of Re(Z44) when increasing frequency and saturation to a constant when beyond 5 GHz suggests a large resistance at very low frequency (due to low open circuit) but fast decay at higher frequency, due to parallel resistors effect from high pass circuit. The proposed mechanism can be simulated by the substrate RC network denoted by solid-line box in Fig.
3.20 and shown in Fig. 3.22 (a) in which Re(Z44) at very low frequency is equivalent to Rbb3
due to open circuit of series Rbb3Cdnw2 but Re(Z44) at very high frequency approach Rbb3//Rbb2
(Rbb3 and Rbb in parallel) due to high pass of Cdnw2. According to the proposed RC network,
44
Z44=Zsub can be derived as given by (3.19). Note that Zsub can be approximated by (3.20) when operating at very low frequency, i.e. 2Cdnw2 2(Rbb2Rbb3)2 1 and then Rbb3 can be extracted from Re(Zsub(LF)) given by (3.21). As for very high frequency, Re(Zsub(HF)) is given by(3.22), which is equal to Rbb3//Rbb2. The experimental data of Re(Zsub(LF)) and Re(Zsub(HF)) can be determined from Fig. 3.21 and then the initial values of Rbb2 and Rbb3 can be extracted from (3.21) and(3.22).
At very low frequency
2 2 2 At very high frequency
2 2 2
45
frequency, according to (3.23)-(3.26).
2 2 At very low frequency,
2 2 with the measured data, as shown in Fig. 3.21.
Considering that the structure of p-well/deep n-well is similar to p-sub/deep n-well, the proposed RC network can be extended to p-well(body)/ deep n-well, by adding body RC network, as shown in Fig. 3.22(b). According to this body RC network defined by dash-line box in Fig. 3.20, the equivalent impedance Zbb‘ can be derived as given by (3.27). Similar with the analysis made on Zsub, Zbb‘(LF) representing Zbb‘ at very low frequency can be
At very low frequency,
2 2 2
1( ) 1
dnw dnw bb
C R R
and 2Cdnw2 1Rdnw(Rdnw Rbb) 1
46
( ) 1
' (1 )
bb LF bb dnw bb
Z R j C R (3.28)
( )
Re(Zbb'LF )Rbb (3.29) At very high frequency,
2 2 2
1( ) 1
dnw dnw bb
C R R
and 2Cdnw2 1Rdnw(Rdnw Rbb) 1
2
( )
1
' ( )
dnw bb bb
bb HF
dnw bb dnw dnw bb
R R j R
Z R R C R R
(3.30)
( )
Re( ' )
( )
dnw bb bb HF
dnw bb
R R
Z R R
(3.31) Unfortunately, Zbb‘ cannot be measured directly and the RC components associated with the body network cannot be extracted simply following (3.28)~(3.31). The solution to treat this problem is to extract the resistances from Re(Y44) under very high frequency, as given by(3.32). Y44 is measured from port-4 (body) with all of the other 3 ports at short state. It can be understood that all of the capacitors become high pass circuits at very high frequency and Re(Y44) can be approximated as two groups of parallel resistors, such as Rbb3//Rbb2 for substrate network and Rbb//Rdnw for body network, as expressed by (3.32). Then, Rdnw can be determined by (3.33) with previously extracted Rbb2 and Rbb3 and Rbb from Re(Y42) or Re(Y43) at very low frequency, as given by (3.3) or (3.4).
Fig. 3.23 outlines the extraction flow for this new body network model. Table 3.3 (a) presents the initial and optimized values of Rbb2 and Rbb3. Table 3.3 (b) summarizes a complete set of the resistances extracted according to the flow in Fig. 3.23 and also the body bias dependence under ZBB, FBB, and RBB. Again, the accuracy of the proposed body network model and the extracted model parameters was verified by a comparison between the measured and simulated Re(Y44) as shown in Fig. 3.24. Note that the simulated results indicate a good match with measurement under various body biases, i.e. ZBB(Vbs=0), FBB (Vbs=0.6V), RBB (Vbs= -0.6V).
47
Fig. 3.19 The cross section of 4T MOSFET with deep n-well tied together with p-well body and connected with port-4.
S
Fig. 3.20 A new body network model proposed for UN65 4-port MOSFET (L65003) in which the deep n-well (DNW) and p-well body (B) are tied together to port-4, and P-sub is connected to ground
48
0 5 10 15 20 25 30 35 40 0
1x103 2x103 3x103 4x103 5x103 6x103
L65003 W2N32 VG=VD=VS=VB=0
Freq (GHz)
measure.ZBB measure.FBB measure.RBB curve fitting.ZBB curve fitting.FBB curve fitting.RBB
Re(Z bb)()
Fig. 3.21 Re(Z44)=Re(Zsub) measured from port-4 (body) with all the other 3 ports (1 : G, 2 : D, 3 : S) at open state and under various body biases : ZBB (Vbs=0), FBB (Vbs=0.6V), and RBB(Vbs=-0.6V).
Body (4)
Rbb2
Cdnw2 Rbb3
P-sub
Intrinsic Body Body(4)
Rbb2
Cdnw2 Rbb3
P-sub
Intrinsic Body
Rbb
Cdnw1 Rdnw
(a) (b)
Body (4)
Rbb2
Cdnw2 Rbb3
P-sub
Intrinsic Body Rbb
Cdnw1 Rdnw
Source (2)
Drain (3)
Cgb1 Cgb2
Gate (3)
(c)
Fig. 3.22 Step by step synthesis of body network model (a) substrate network for deep
49
n-well/p-substrate (b) p-well body network for p-well/deep n-well and substrate for deep n-well/p-substrate (c) a complete body network model for L65003 4T RFMOSFET
Fig. 3.23 A new body network model parameters extraction flow for 4-port RF MOSFET with equivalent circuit shown in Fig. 3.20
Table 3.3 Resistance parameters extracted for the new body network model (a) initial and optimized Rbb2 and Rbb3
Zsub default optimization
Rbb2 384 664
Rbb3 6741.12 5484
(b) Rbb, Rdnw Rbb2 and Rbb3 after optimization
Optimized parameters ZBB (Vbs=0V) FBB (Vbs=0.6V) RBB(Vbs=-0.6V)
Rbb () 958 977 842
Rdnw ( 476 202 233
Rbb2 () 664 874 813
Rbb3 () 5484 7994 8192
50
0 5 10 15 20 25 30 35 40 0.0
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
5.5 UN65-L65003 4T MOSFET VG=V
D=V
S=0
Meas Model
FBB
RBB
ZBB
Re ( Y
44) ( 10
-3)
Freq (GHz)
Fig. 3.24 Comparison of measured and simulated Re(Y44) using the new body network model and extracted parameters under ZBB (Vbs=0), FBB (Vbs=0.6V), and RBB(Vbs=-0.6V).
The final step to complete the extraction flow is the extraction of gate to body capacitances, namely Cgb1 and Cgb2 as shown in Fig. 3.20. Then the step by step synthesis of the body network model is moved from Fig. 3.22(b) to Fig. 3.22(c). In general, gate to body capacitance can be extracted from -Im(Y14)/ as shown in Fig. 3.25. However, the strong frequency dependence revealed in Fig. 3.25 with a fast fall off in lower frequency region and then saturation to a constant at frequency beyond 25GHz suggests that the gate to body capacitance is composed frequency independent and frequency dependent components, denoted as Cgb1 and Cgb2, respectively. The origin responsible for Cgb1 and Cgb2 can be explained by RF MOSFET layout shown in Fig. 3.26. The frequency independent component (Cgb1) is contributed from gate contact/metal to body contact/metal coupling capacitance. The frequency dependent component (Cgb2) comes from gate to channel (body) coupling, which may reveal non-quasi-static effect. According to the measured -Im(Y14)/ shown in Fig. 3.25 and the analysis supported by Fig. 3.26 , Cgb1 and Cgb2 can be extracted as follows, given by (3.34) and(3.35).
51
14(HF) 25
1
Im(Y ) |f GHz Cgb
(3.34)
14(LF) 1
2 1
Im(Y ) |f GHz
gb gb
C C
(3.35)
0 5 10 15 20 25 30 35 40 2.0
2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
C
gb=-Im(Y
14)/
( fF )
(c)
Freq-dependent, Cgb2
Freq-independent Cgb1
Freq (GHz)
Cold device Vg=Vd=Vs=Vb=0 Cgb= -Im(Y
14)/
Fig. 3.25 The gate to body capacitances measured from 4-port Y-parameters after openM3 deembedding on 4-port RF MOSFET –Im(Y14)/=Cgb=Cgb1+Cgb2
B
G
Fig. 3.26 RF MOSFET layout remarked with poly gate fingers, gate contact and metal to contacts, body contacts and metal to contacts. The metals to gate contacts and body contacts will contribute inter-metal coupling capacitance.
52