Drain Current Noise Spectrum Measurement in 0.18 ?m MOSFET Using Integrated SiGe HBT Low-Noise Transimpedance Amplifier by Jingshan Wang A thesis submitted to the Graduate Faculty of Auburn University in partial fulfillment of the requirements for the Degree of Master of Science Auburn, Alabama August 3, 2013 Keywords: Drain Current Noise, Thermal Noise, 1/f noise, Corner Frequency, TIA, S-Parameter Copyright 2013 by Jingshan Wang Approved by Guofu Niu, Chair, Alumni Professor of Electrical and Computer Engineering Fa Foster Dai, Professor of Electrical and Computer Engineering Bogdan Wilamowski, Professor of Electrical and Computer Engineering ii Abstract We measured drain current noise power spectral density (PSD) in 0.18 ?m metal oxide semiconductor field effect transistor (MOSFET) using integrated Silicon-Germanium (SiGe) Heterojunction Bipolar Transistor (HBT) low-noise transimpedance amplifier (TIA). This measurement system extends the noise-measuring capabilities beyond 2.5 GHz to detect the white noise component beyond the 1/f noise corner frequency. In this work, the corner frequency is approximately 2 GHz, which comes from radio frequency (RF) thermal noise measurement directly instead of the extension line of 1/f noise at low frequencies. PSD of drain current thermal noise is in the range from 1 22-10? A2/Hz to 5 22-10? A2/Hz for drain current from 1mA to 12 mA. S-parameters are measured to calculate the gain of device under test (DUT) and TIA system. iii Acknowledgments First of all, I would like to express my deepest gratitude to my faculty advisor Dr. Guofu Niu for his technical and moral support throughout my Master?s program, without which this thesis would not have been possible. His vast knowledge and valuable feedback greatly contribute to the work. His patience and constant encouragement strengthened me during my whole graduate study. Several people deserve special recognition for their contributions to my thesis. I would like thank Dr. Bogdan Wilamowski and Dr. Fa Dai for serving on my committee and all the help throughout my graduate study. This work would not have materialized without Dr. Fa Dai?s support of measurement equipment. Also, I am very grateful to Dr. Stewart Taylor and Dr. Xiaoyun Wei for sharing their TIA design. Many thanks also go to Dr. Stuart Wentworth and Dr. Lloyd Riggs. Their recommendation and encouragement help me gain both the opportunity to continue my academic research and the confidence to face new challenge in the future. Besides, I would like to thank my group mates Hongyun Xie, Lan Luo, Ziyan Xu, Zhen Li, Ruocan Wang and Xiaojia Jia. The time on and off working with them was pleasant and memorable. Finally, I would like to thank my family, my mother, my brother, my sister in law, my parents in law and my husband Qiang Gu, for their selfless support and love, without them, I iv would not have begun my Master?s program. Most importantly, I need to thank God for the most precious gift I ever have, Derrick, my one year old son. He helps me realize the responsibility I have for whatever I am working on. v Table of Contents Abstract ......................................................................................................................................... ii Acknowledgments........................................................................................................................ iii List of Figures ............................................................................................................................. vii List of Abbreviations .................................................................................................................... x Chapter 1 Introduction .................................................................................................................. 1 1.1 Thermal Noise ............................................................................................................ 3 1.1.1 Drain Current Thermal Noise ............................................................................ 3 1.1.2 Induced Gate Noise and Gate-Drain Correlation Noise .................................... 5 1.2 Flicker Noise (1/f Noise) ............................................................................................ 7 Chapter 2 Drain Current Noise Measurement Methods ............................................................. 10 2.1 Noise Parameters Measurement System ................................................................... 10 2.2 Noise PSD Measurement System Using Discrete TIA ............................................. 14 2.3 Noise PSD Measurement System Using Integrated TIA on Chip ............................ 16 2.3.1 Noise Measurement System in the Thesis ....................................................... 17 2.3.2 SiGe HBT TIA ................................................................................................. 28 Chapter 3 Noise Measurement and Data Analysis .................................................................... 33 3.1 Noise Measurement .................................................................................................. 33 3.2 Experimental Results Analysis ................................................................................. 37 vi Appendix Appendix A On-Wafer DC Measurements ................................................................................ 46 A.1 Introduction ............................................................................................................. 46 A.2 DC Measurements for Standalone MOSFET .......................................................... 48 A.2.1 IDS-VDS and IDS-VGS Measurements ................................................................ 48 A.2.2 gds and gm ....................................................................................................... 49 A.3 DC Measurements for Integrated DUT in Circuit ................................................... 51 A.3.1 Measurement Equivalent Circuit .................................................................... 51 A.3.2 IDS-VDS Measurement and gm Extraction ......................................................... 52 Appendix B On-Wafer S-Parameters Measurements ................................................................ 57 B.1 Introduction ............................................................................................................. 57 B.2 S-Parameters Measurements System ...................................................................... 60 B.3 S-Parameters Measurements for Standalone MOSFET ........................................... 63 B.4 S-Parameters Measurements for Integrated DUT and TIA System ......................... 67 Bibliography ............................................................................................................................. 72 vii List of Figures Figure 1.1 A typical plot of the drain current noise versus frequency in log-log axes ................. 2 Figure 1.2 Measured ? values for different technologies reported in the literature ...................... 5 Figure 1.3 Extracted channel noise, induced gate noise, correlation noise, and cross-correlation coefficient as a function of frequencies for devices with different channel lengths ... 7 Figure 2.1 System configuration for radio frequency noise measurements................................ 11 Figure 2.2 Schematic diagram for the measurement system shown in Figure 2.1 ..................... 12 Figure 2.3 Block diagram of the noise PSD measurement setup ................................................ 14 Figure 2.4 A typical equivalent input noise VSD ....................................................................... 15 Figure 2.5 Noise measurement system using integrated TIA on chip ........................................ 17 Figure 2.6 A sample noise power measured result ..................................................................... 17 Figure 2.7 Noise measurement system using SiGe TIA on chip ................................................ 18 Figure 2.8 Die photo with DUT and TIA on chip....................................................................... 19 Figure 2.9 Simple equivalent circuit for noise from Rg,bias ......................................................... 20 Figure 2.10 A typical output power of amplified noise .............................................................. 21 Figure 2.11 A typical output power spectral density of output power........................................ 22 Figure 2.12 A typical output VSD .............................................................................................. 23 Figure 2.13 Noise measurement system equivalent circuit ........................................................ 24 Figure 2.14 Equivalent two ports network .................................................................................. 24 Figure 2.15 A typical noiseinV, vs frequency .................................................................................. 26 viii Figure 2.16 A typical Sid vs frequency ....................................................................................... 27 Figure 2.17 Noise factor ? versus frequency .............................................................................. 27 Figure 2.18 A typical output PSD of NMOS + TIA system, TIA noise floor and SA noise floor versus frequency ..................................................................................................... 29 Figure 2.19 Power spectral density of output power with/without TIA noise effect .................. 30 Figure 2.20 Measured I-V curve of TIA with DUT biased off ................................................... 31 Figure 2.21 A typical totalZ vs frequency ................................................................................... 31 Figure 2.22 A typical Stia and Sid vs frequency. .......................................................................... 32 Figure 3.1a Drain current noise measurement system setup ....................................................... 34 Figure 3.1b Drain current noise measurement system diagram .................................................. 34 Figure 3.2 Block diagram of operation theory for spectrum analyzer ........................................ 35 Figure 3.3 Die photo with right AC probe and DC probes on pads............................................ 36 Figure 3.4 Configuration of Eye-Pass DC probes (Cascade Microtech Company) ................... 37 Figure 3.5 Drain current noise spectral density Sid and noise factor ? versus VDS ..................... 38 Figure 3.6 Noise factor ? with different biases ........................................................................... 38 Figure 3.7 Extracted noise factor ? vs VGS in 0.18 ?m technology ............................................ 39 Figure 3.8 Drain current noise PSD Sid vs Vgs ........................................................................... 39 Figure 3.9 Drain current noise PSD Sid vs Vgs in 0.18 ?m technology ....................................... 40 Figure 3.10 Drain current noise PSD Sid vs drain current IDS .................................................... 40 Figure 3.11 Drain current noise PSD Sid vs drain current IDS in 90 nm technology ................... 41 Figure 3.12 Comparison of output noise power ........................................................................ 42 Figure 3.13 Comparison of Sid .................................................................................................... 43 Figure 3.14 Comparison of ? ...................................................................................................... 43 Figure 3.15 Sid vs frequency at fixed VGS ................................................................................... 44 ix Figure 3.16 Sid vs frequency at fixed VDS ................................................................................... 45 x List of Abbreviations SiGe Silicon-Germanium HBT Heterojunction Bipolar Transistor TIA Transimpedance Amplifier DUT Device Under Test MOSFET Metal?Oxide?Semiconductor Field-Effect Transistor NMOS N-Type Channel Metal?Oxide?Semiconductor PSD Power Spectral Density RF Radio Frequency LNA Low Noise Amplifier SA Spectrum Analyzer VSD Voltage Spectral Density NF Noise Figure 1 Chapter 1 Introduction All semiconductor devices generate noise that can interfere with weak signals when used in circuits. Therefore, it is important to find ways to measure noise. There are several types of fundamental noise present in semiconductor device: thermal noise, flicker noise also called 1/f noise, shot noise, generation-recombination noise and burst noise. These noises vary with device structure and operating conditions. Thermal noise and flicker noise are two major types of noise in metal oxide semiconductor field effect transistor (MOSFET). The purpose of Chapter 1 is to introduce these different types of noise, especially the thermal noise in MOSFET. A typical drain current noise PSD versus frequency plot is shown in Figure 1.1.Theoretically, a plot of drain current noise power-spectral density versus frequency in log-log axes is a straight line with the slope of approximately 1/f at low frequency and a horizontal line at high frequency. There is a corner at the intersection of these two lines which is valued from several hertz to several gigahertz depending on device geometry, construction, and bias. With the development of semiconductor technology, the corner frequency becomes higher and higher [1]. Usually, 1/f noise is measured using time domain equipment [2], such as dynamic signal analyzer. Since the drain current thermal noise is too weak to be measured directly, the main method for thermal noise measurement is based on measurement of noise parameters [3], such as noise figure (NF). Chapter 2 gives background material for thermal noise measurement method. The history of thermal noise measurement for MOSFET is introduced. Our noise measurement system and basic working theory are described. The algorithm used here is specific for our test system. The on-chip Silicon-Germanium (SiGe) Heterojunction Bipolar Transistor (HBT) 2 Transimpedance Amplifier (TIA) used to amplify drain current noise, is introduced. There are several requirements for TIA to meet to be able to measure MOSFET drain current noise. Chapter 3 describes the process of noise measurement and analysis of experimental result. The power spectral density (PSD) of drain current noise Sid and thermal noise factor ? are extracted. We find that Sid increases with IDS at fixed VDS. It also increases with VDS at fixed VGS as well as increases with VGS at fixed VDS. The values of noise factor ? from our extraction are consistent with reported values in literature for similar technologies. Figure 1.1 A typical plot of the drain current noise versus frequency in log-log axes. Different types of fundamental noise are present in semiconductor devices: thermal noise, flicker noise (1/f noise), generation-recombination noise, shot noise and burst noise. Thermal noise and shot noise are white noise. 1/f noise origin remains a mystery. Thermal noise and 1/f noise are the two major concerns for drain current noise measurement. With channel length scaling and the carrier mobility improvement, the cut-off frequency of MOSFET can go up to hundreds of gigahertz [4]. Hence, MOSFET are widely used in high 3 speed radio frequency RF integrated circuits. In these circuits, drain current noise becomes a critical issue. 1.1 Thermal Noise Thermal noise is generated by random motion of free carriers in resistive materials. It is present in all circuit elements containing resistance regardless of any applied voltage. In 1928, J.B. Johnson was the first to prove that there is fluctuating movement of charges in thermal equilibrium [5]. In an ideal resistor, thermal noise is independent of frequency. In other words, the power spectral density is nearly constant throughout the frequency spectrum for a fixed bandwidth. 1.1.1 Drain Current Thermal Noise Starting from van der Ziel (1962) [6], many drain current thermal noise models are developed. In 2002, Chen and Deen proposed their new model which considered the channel length modulation (CLM) effect [7]. In 2005, Paasschens, Scholten & van Langevelde provided channel thermal noise model which considered velocity saturation effect and separated the position and voltage dependence for the channel conductance [8]. The noise factor ? was discussed in both [7] and [8]. This noise factor was introduced in one of the most prevalent models for drain current noise PSD, Sid, in long channel device at strong inversion region by van der Ziel [1]: 0id 4S dsgKT ?? ? , (1.1) where )('g 0ds TGSoxDSDS VVLWCVI ????? ? , (1.2) 4 ? ?? Vd 0 000 Vd 0 0 200 ]/)([ ]/)([ dVgVg dVgVg? , ? 1.3? K is Boltzmann constant, T is absolute temperature, ? is thermal noise factor, 0dsg is drain- source conductance evaluated at drain-source voltage DSV = 0 V, W is channel width, L is channel length, ? is channel mobility, oxC' is oxide capacitance per unit area, GSV is gate-source voltage and TV is threshold voltage, g(x) is the conductance per unit length, V0(x) is the channel potential at the point, g0 is conductance at V0=0. The value of ? is 2/3 and 1 in the saturation and triode regions for long channel device, respectively. With technology scaling, oxC' is increased, ? is decreased while the product ? oxC' is increased [4]. If LW and TGS VV ? are fixed, both 0dsg and Sid increase with reducing the physical size of devices. If L is fixed for a certain technology, increasing W can result in increasing Sid. The parameter ? as a noise factor is defined [1] from equation (1.1) as 0 id4 S dsKTg?? . (1.4) Later, the ? values are found higher than 2/3 in a short channel transistors working in the saturation region [7-10]. Actually, ? values are different with different semiconductor technologies [10]. ? is widely used in literatures to demonstrate the enhanced channel thermal noise in short channel transistors as shown in Figure 1.2. The ? range measured in our work is from 0.22 to 0.53 at bias sets VDS from 0.18 V to 1 V and VGS from 0.51 V to 0.64 V in 0.18 ?m technology. The bias sets we used are in moderate region. The values of ? are from 1 to 2 at bias sets VDS from 1.5V to 1.8V and VGS from 1 to 1.8 in the same technology shown in Figure 1.2. More comparisons are discussed in section 3.2. 5 Figure 1.2 Measured ? values for different technologies reported in the literature [10]. 1.1.2 Induced Gate Noise Random fluctuations of the potential in the channel are coupled to the gate terminal through the oxide capacitance leading to induced potential fluctuations on the gate. Drain noise and gate noise are correlated in MOSFET with frequency dependence. In van der Ziel model [1], PSD of gate current noise Sig is given by: GgKT?4Sig ? (1.5) where? = 4/3, and the saturation gate conductance Gg is: m oxG g WLC 22 )'(454g ?? (1.6) 6 where gm is transconductance. In saturation, the cross-correlation of reduced gate current noise and drain current noise is given by [1], KT qWLCii oxd 4'i *g ?? (1.7) where i is the imaginary unit. The correlation coefficient between gate and drain current fluctuations is i0.4 [1]. The gate-to-source capacitance decreases when the channel length decreases. Hence, both induced gate noise and its correlation with the channel thermal noise decrease [11]. The origin of this comparatively low coefficient is induced charge profile on the gate, and it will not occur in short channel MOSFET [12]. In 0.18 ?m technology which is used in our project, PSDs of the channel noise (drain current noise), induced gate current noise, correlation of induced gate current noise and drain current noise, and cross-correlation coefficient are extracted [12] and shown in Figure 1.3. 7 Figure 1.3 Channel noise, induced gate noise, correlation of these two noises, and cross- correlation coefficient as a function of frequencies for devices with different channel lengths [12]. 1.2 Flicker Noise (1/f Noise) Unlike other noises, there are many different explanations of the origin of flicker noise or 1/f noise. There are two major models and concepts for 1/f noise [13]. The first theory describes the origin of 1/f as the random fluctuation of the number of carriers in the channel because of the fluctuations in the surface potential [4]. PSD of the equivalent drain noise current Sid can be calculated by [4] 2m21 g11'S cfWLC K oxid ? (1.8) 8 where )('g m TGSox VVCLW ?? ? (1.9) in saturation region. ?1 is bias dependent quantity which increases with technology scaling, c is between 0.7~1.2 for n-channel device. This model obtains 1/f noise by superposing many different spectra of generation-recombination noise with a specific statistical distribution of [13]. Free carriers are randomly trapped and released by traps located near the silicon-oxide interface, causing noise in drop current. At corner frequency, thermal noise Sid from equation (1.1) equals 1/f noise Sid from equation (1.8), so the corner frequency, cornerf , is calculated by WLC gK gKTf oxdsc o r n e r 2 21 0 '4 1 m?? (1.10) Substituting equation (1.2) and (1.9) to (1.10), cornerf can be simplified to ?? KTLKf c o r n e r 4 )V-V(2 TG1? . (1.11) If (VG-VT) is fixed, cornerf increases with decreasing L. With technology scaling, cornerf becomes bigger and bigger. The second theory attributes 1/f noise to the mobility [4]. Under this theory, PSD of the equivalent drain noise current is given by 2GSd 11'VS m oxi gfWLCK ??? (1.12) where ?(???) is bias dependent quantity. In this model, both lattice scattering and impurity scattering are considered. It is assumed that only scattering on the silicon lattice generates 1/f noise. Similar to the first model, if VG and VT are fixed, cornerf increases with decreasing L. 9 In our work, DUT is a NMOS with channel width 2.3100W ?? ?m where100 is gate finger number, and channel length L = 0.18 ?m. The reported noise factor ? for that technology is approximately in the range from 2/3 to 2 for several bias sets in [7], [12] and [14]. The corner frequency we measured is approximately 2 GHz. 10 Chapter 2 Drain Current Noise Measurement Methods As mentioned in Chapter 1, the corner frequency becomes higher and higher with the development of semiconductor technology in MOSFET. To measure thermal noise which dominates after 1/f corner frequency, the noise measurement should be taken in very high frequency and it is up to several gigahertz for modern semiconductor technologies. Another difficulty is the thermal noise is too weak to be measured directly on noise measurement equipment. The main method for thermal noise measurement is based on measurement of noise parameters [3], F or NF. From 1986, TIA is used as a low noise amplifier (LNA) to boost the drain current noise and improve the measurement accuracy [8]. Approximately one decade later, TIA is integrated with DUT on chip for gallium arsenide GaAs metal-semiconductor field effect transistor MESFET[15] to reduce the parasitic capacitance and inductance. In our project, SiGe HBT TIA is integrated with a NMOS on chip to help with drain current noise measurement in 0.18 ?m technology. 2.1 Noise-Parameters Measurement System Different from the 1/f noise measurement, where the noise PSD can be directly measured using a dynamic signal analyzer, the main thermal noise measurement method is measuring the noise factor F (or noise figure NF in dB) and/or its well-known noise parameters developed by Haus et al. in [16], to evaluate the thermal noise characteristics. In [16], F (or NF) is presented as a mathematical equation 11 s n2optsm in GR|Y-Y|FNF ?? N (2.1) where Ys = Gs + i?B s is the admittance of source, Yopt = Gopt + i?B opt is optimized source admittance, NFmin is minimum noise figure, Rn is equivalent noise resistance. This representation is based on a noisy two-port network expended from Rothe and Dahlke in [17]. Haus et al.?s impedance-based representation [16] demonstrates the dependence of noise factors on the source admittances attached to the input port of the noisy two-port network. In the measurement, NF is measured by Noise Figure Analyzer NFA for certain Ys firstly, noise parameters NFmin, Rn and Yopt can be calculated from equation (2.1). Then these noise parameters are expressed as functions of two-port network chain representation A, B, C and D [16]. Finally, drain current noise can be acquired [18-22]. Under the theory of this two-port noise representation [16], many noise measurement and extraction methods have been developed [18-22]. Figure 2.1 System configuration for radio frequency noise measurements [22]. 12 Figure 2.1 is the system configuration for noise measurement in [22]. This system consists of a noise source, a noise figure analyzer (NFA), a vector network analyzer (VNA), low noise amplifier (LNA), microwave impedance tuners, power supply and other peripheral components, such as PC, switches and bias tees. In Y-factor or hot/cold-source technique [23], the calibrated noise source generates two noise outputs with different equivalent noise temperatures, hot temperature (Thot) and cold temperature (Tcold). Under the theory of impedance- based two-port noise network [16], the impedance of the network should adjust. The source tuner and load tuner are used to provide different source admittances and to match the output of the DUT for a maximum power transfer, respectively [23]. These two tuners are controlled by tuner controller in this system. LNA is used to boost the weak noise signal to allow the noise signal been measured by NFA. PNA is to measure S-parameters of the impedance tuner ST and receiver SR which are defined in the following section and shown in Figure 2.2. Figure 2.2 Schematic diagram for the measurement system shown in Figure 2.1[22]. Figure 2.2 is the schematic diagram for the measurement system shown in Figure 2.1. This measurement system contains three parts: noise source, an impedance tuner and a receiver. In the system calibration stage, a THRU line is placed between the input and output probes. Then, 13 noise reference plane in Figure 2.2 is corresponding to plane B in Figure 2.1. In other words, the impedance tuner in Figure 2.2 contains all the components from tuner reference plane to plane B in Figure 2.1, and the receiver in Figure 2.2 includes LNA, NFA and the cable between them in Figure 2.1. In the measurement stage, DUT takes the place of the THRU line. The noise reference plane in Figure 2.2 is moved to plane A in Figure 2.1. All the components between plane A and plane B in Figure 2.1 are included in receiver in Figure 2.2. Based on the noise reference plane in Figure 2.2, the noise power Pn detected by NFA is expressed by [22] )]22||||1(||||||4[ 4GP 22222tr sc o rsc o rsc o rsuns e fsn XBRGZYuZif R sfKTR ????????? (2.2) where ? f is noise bandwidth, Rs is source resistance, Xs is source reactance, Zs is source impedance seen at the noise reference plane (= Rs + i ? Xs), u is input referred noise voltage [17], uni is input referred noise current [17], Gcor is correlation conductance, Bcor is correlation susceptance, Ycor is complex correlation admittance (= Gcor + i ?B cor) [17], Gtr is transducer power gain of the receiver, Tseff is effective source temperature experienced at the noise reference plane. After conversions step by step in [22], noise parameters NFmin, Rn and Yopt are expressed using the parameters shown in equation (2.2). Then NFmin, Rn and Yopt are represent by chain parameters A, B, C and D. At last, the drain current noise can be calculated. As the main high frequency measurement method, the noise parameters measurement method is widely used and there are lots of relative literatures [18-23]. However, drain current noise cannot be measured directly using this method. Another method which can measured the drain current noise directly is described in section 2.2. 14 2.2 Noise PSD Measurement Using Discrete TIA Another method of noise measurement is to amplify drain current noise using a low-noise transimpedance amplifier TIA [8,22,24]. Figure 2.3 shows the setup used by Tedja in measuring noise of spectrum of MOSFET from a 1.2 ?m technology. Figure 2.3 Block diagram of the noise PSD measurement set-up [24]. The noise current at the drain node of the DUT (or equivalently the input node of TIA) is dominated by the drain current noise. In the noise measurement stage, switch S is open and the drain current noise flows into a discrete TIA. Then the combined voltage from drain current noise and TIA noise at the output of the TIA is further amplified by a gain stage. At last, the resulting voltage was detected by SA. In [24], the measured result on SA is considered to be the drain current noise PSD multiplied by the gain of the amplifiers following the DUT if the TIA and the gain stage were noiseless. The extra noise in the measurement system, such as the noise from TIA, gain stage and biasing circuit, is re-measured when the DUT is turned off. In the system transfer function measurement stage, switch S is closed and a known signal from SA is fed into the DUT. After subtracting the extra noise from the output noise voltage spectral density 15 VSD, the output noise VSD is referred to the input of the DUT which is called input referred noise VSD (or called gate referred noise VSD in [25]) by dividing it by the overall gain or transfer function of the whole noise measurement system. A typical input noise voltage spectral density is shown in Figure 2.4 Figure 2.4 A typical equivalent input noise VSD [24]. The equivalent PSD of input voltage noise is given by [24] )'(414V 2 GRRKTf BgKTf B m in ????? ?? (2.2) where m dsgg 0??? , B m mbB RggR 22'? , B is the flicker noise coefficient, ? is the power of the 1/f noise term (?1), 'BR is the effective bulk noise resistance, BR is the bulk resistance, GR is the resistance of the poly-silicon gate and the gate interconnects. This equation also implies the gate is AC shorted to ground during noise measurement, which necessitates consideration of BR and GR noise as )'(4 GRRKT B ? . Sid can be calculated by multiplying Vin 2 by gm2. 16 One disadvantage of the noise measurement system in [24] is that the 50 ? source resistance of SA is not considered in the measured overall gain or transfer function. At higher frequencies, the 50 ? source resistance of SA cannot be neglected, as we will illustrate below in section 2.3. In our work, the overall gain portion is optimized by using S-parameters measurement instead of the transfer function measurement in [24]. 2.3 Noise PSD Measurement System Using Integrated TIA on Chip To reduce the parasitic capacitance and inductance, the noise PSD measurement system is optimized by integrating the TIA with DUT on chip [15]. In [15], both GaAs MESFET DUT and TIA are integrated on chip as shown in Figure 2.5. The frequency range for this system is up to 10 MHz and a sample measured result is shown in Figure 2.6. One contribution of our work is showing the measured data in much higher frequency up to approximate 2.5 GHz in 0.18 ?m technology. Figure 2.5 Noise measurement system using integrated TIA on chip. 17 Figure 2.6 A sample noise power measured result [5]. 2.3.1 Noise Measurement System in the Thesis SiGe HBTs are widely used to design low noise amplifier [26]. In our project, we use a new TIA which has a bandwidth up to 2.5 GHz to allow us to detect the white noise component beyond the 1/f noise corner frequency. As shown in Figure 2.7, DUT and TIA are AC coupled and the output impedance of the DUT plus resistor on drain node is much larger than the equivalent input impedance of Cc plus TIA part, so that the drain current noise current id can flow into the TIA and be amplified there. A voltage buffer with an output impedance of 50 ? is used to drive a 50 ? spectrum analyzer. The size of DUT is: W/L= 100 ? 3.2 / 0.18 ?m, where 100 is the number of gate figures, 3.2 ?m is the gate finger width. As discussed in section 1.2, the chosen size of DUT affects noise measurement. With increasing the channel width, the thermal noise increases while 1/f noise 18 decreases. Hence the corner frequency decreases, making the thermal noise easier to measure. However, the channel width cannot go to infinity. One important reason is that the output impedance of DUT decreases with increasing the channel width. Figure 2.7 Noise measurement system using SiGe TIA on chip. Figure 2.8 Die photo with DUT and TIA on chip. 19 Figure 2.8 shows one test die under microscope. The big circle on the right contains the DUT and TIA in circuit. The circle on the left contains the standalone NMOS which is designed to be the same structure as the NMOS in the circuit, and the size of the NMOS is the same as DUT in circuit. The standalone NMOS is used for DC and RF characteristics. In Figure 2.7, DC pad VCC is used to connect DC power supply for TIA. DC pad VDD is used to connect DC power for drain terminal of the DUT. There is a resistor RDD between VDD and drain terminal to avoid drain current noise flow out from VDD pad. Because that RDD produces noise in, RDD = DDRKT /4 , to ignore in, RDD when measuring drain current noise id= 04 dsgKT? , in, RDD should much less than id, which means 01 dsDD gR ??? . ? is around 1, so 0 1 dsDD gR ?? . In our work, 0dsg is in the range of 30 mS to 80 mS, so 0 1 dsg is approximately from 10? to 30 ?. Hence, 2 K? is a big enough value for DDR . Since there is a resistor RDD between VDD and drain terminal, the value of VDD is much higher than the voltage applied to drain terminal. If DDR is too large, VDD need to be very high to achieve a certain VDS, which may exceed the range of DC power supply. A tiny current source iDS, such as10 nA, is applied to DC pad VDS to monitor the voltage actually applied to drain terminal VDS. DC pad VGS or AC pad IN is used to connect power supply for the gate. There is a 20 K? resistor Rg,bias between VGS pad and the gate terminal. 20 Figure 2.9 Simplified equivalent circuit for noise from Rg,bias. Consider the noise from Rg,bias as an equivalent voltage noise source VRg,bias in Figure 2.9, the PSD of drain current noise from Rg,bias Rgbiasid,S can be calculated as 2 , 22 2 , 22R g b ia sid , ,Rm,RmS ? ???????????????? ?? b ia sg in b ia sgin in R ZVgRZ ZVg b ia sgb ia sg (2.3) where Zin is equivalent input impedance approximately 100 ?, gm is the transconductance of the DUT, and the square of equivalent input voltage for Rgid,S , 2 ,RgbiasinV , is given by n o is egn o is eg in R ZKTR ZK T RV inb ia sg R g b ia sin , 22 , 2 44 , , ???? ??? ? ?? (2.4) To ignore Rgbiasid,S , 2 ,RgbiasinV in equation (2.4) should be much smaller than 204 mg gKT ds? . 20 mg gds and inZ can be acquired by DC measurement and S-parameters measurement which are discussed in detail in Appendix A and B, respectively. So we can choose a large enough value for Rg,bias. 21 Hence, the output power show on SA is the amplified drain current noise plus noise power of TIA (TIA noise floor). The TIA noise contribution to the output is first measured by turning the device off with VGS and VDD both set to 0 (never set VGS=0 and use high VDD, or device is damaged). VGS and VDD are ramped up gradually by monitoring VDS and IDS to desired value, noise measurement is made by SA. Subtracting the previously measured TIA noise contribution gives the desired output noise contributed by DUT. The output power data saved from SA is in dBm unit. To remove the TIA noise, all the measured powers which include TIA noise power and total output power data are convert to watt unit first, and then the TIA noise power is removed from all the total output power leaving the amplified device noise power. After subtracting the TIA noise, the output power is converted back to dBm unit. An example is shown for an active bias point of VGS = 0.51V and VDS = 1V in Figure 2.10. Figure 2.10 A typical output power of amplified noise. 22 The output power decreases from -78 dBm at 100 MHz to -92 dBm at 3 GHz. noiseoN, , PSD of the output power, is then calculated and expressed in unit of dBm/Hz as follows [27, 28]: BWP d B mout 1010n o is eo, l o g10|BWmW1 Pl o g10N ???? (2.5) where BW is measurement bandwidth, and is determined by resolution bandwidth RBW in SA settings [29]. The voltage spectral density VSD of amplified drain current noise noiseoV, (V/ Hz ) by converting power to voltage which is given by on o is eNon o is eo Z?? ? 310,, 10V (2.6) where Zo = 50 ? is the SA input impedance. Figure 2.11 and 2.12 are noiseoN, and noiseoV, for the same bias set VGS = 0.51V and VDS = 1V as Figure 2.10. Figure 2.11 A typical output power spectral density of output power. 23 Figure 2.12 A typical output VSD. Consider the noise measurement system in Figure 2.7 as an equivalent input referred noise voltage source noiseinV, added to a noiseless circuit as shown in Figure 2.13. noiseoV, has been acquired above. The drain current noise id related to noiseoV, by totalnoiseo ZidV ??, (2.7) where b u fferTIAto ta l AZZ ?? , TIAZ is TIA gain, bufferA is buffer voltage gain. To get totalZ , an input voltage signal Vin which is large enough to ignore id is added to input port of the network shown in Figure 2.13. The output voltage of the test signal Vout is given by to ta lminout ZgVV ??? . (2.8) Using equivalent (2.7) and (2.8), an input referred noise voltage noiseinV, can be defined as in out noiseo noisein V V VV , , ? . (2.9) 24 Such that noiseinmVg ,id ? . So the next question is to find in outVV . In [24], a spectrum analyzer is used. This however cannot account for the 50 ? source resistance, which is not negligible for high frequency, where Zin is only 100 ? in Figure B.17. We solve this problem using S- parameters measurement. The S-parameters can be converted to Y-parameters which we can use to find in outVV as below. Figure 2.13 Noise measurement system equivalent circuit. Figure 2.14 Equivalent two ports network. 25 Consider a two port network as shown in Figure 2.14, and apply an input test voltage Vin with a Zo load, the current at port 2 I2 is given by [30] 1212222 YY VVI ?? (2.10) where V1 and V2 are the voltages at port 1 and 2, respectively. Y-parameters Y21 and Y22 can be calculated from S-parameters by [27] 21122211 21 o12 SSS1S1 2S-Z1Y ????? ??? (2.11a) 21122211 21122211 o22 SSS1S1 SSS1S1Z1Y ????? ???? ?? ???? (2.11b) Substituting oZ VI 22 -? , outV = 2V and inV = 1V into equation (2.10), in outVV is obtained as o o in out ZZVVVV 22 21 1 2 Y1 Y???? . (2.12) Using equation (2.9) to (2.12), we can calculate noiseinV, from noiseoV, and S-parameters. noiseoV, has been acquired from noise spectrum measurement discussed above and standard on-wafer S- parameters measurement is discussed in Appendix B. noiseinV, (V/ Hz ) at bias VGS = 0.51V and VDS = 1V is plotted in Figure 2.15, noiseinV, decreases from 100 MHz to 2 GHz, and then it becomes independent of frequency from 2 GHz to 3 GHz. 26 Figure 2.15 A typical noiseinV, ( V/ Hz ) vs frequency. The drain current noise PSD Sid can now be calculated from gate-referred noise voltage noiseinV, by 2,inid )(S n o isemVg? . (2.13) Sid at VGS = 0.51V and VDS = 1V is shown in Figure 2.16. 27 Figure 2.16 A typical Sid (A2/Hz) vs freqency . From Figure 2.16, we can see the corner frequency for our DUT is around 2 GHz. Sid at lower frequencies is nearly proportional to 1/f. Sid of the thermal noise component beyond 1/f corner frequency is 1.8?10-22 A2/Hz. When frequency is higher than 2.5 GHz, Sid increases a little near 3 GHz. The possible reason is the poor performance of TIA in that frequency range which is discussed in section 3.2.2. Since noise factor ? is widely used in literatures to demonstrate the enhanced channel thermal noise in short channel transistors. After extracting Sid, the noise factor ? is derived using equation (1.3). For bias set VGS = 0.51V and VDS = 1V, ? is shown in Figure 2.17. 28 Figure 2.17 Noise factor ? versus frequency. 2.3.2 SiGe HBT TIA To use TIA as the low noise amplifier, the system must satisfy two requirements to make sure the device noise can be resolved from other noise [8] because that the device noise is not the only noise source in the measurement system. The first requirement is that the amplified noise of the device must be greater than the input noise of the spectrum analyzer which is the SA noise floor. The second requirement is that the equivalent input noise of the amplifier which is the TIA noise floor must be less than the device noise. In both triode and saturation regions of the MOSFET, the noise amplifier should be driven in a large resistances region presented by the DUT. In the saturation region of MOSFET, MOSFET works like a high resistance current source at output terminal. This phenomenon requires that the amplifier following the MOSFET has a low equivalent input current noise. In our work, SA noise floor is the noise measurement result for only SA and the cable between output probe and SA. TIA noise floor is the noise measurement result when DUT turned off. NMOS device (DUT) amplified noise is measured when both DUT and TIA are biased. 29 Figure 2.18 A typical output PSD of NMOS + TIA system, TIA noise floor and SA noise floor versus frequency As shown in Figure 2.18, the TIA satisfies all the design requirements described above. Comparing the NMOS biased and SA noise floor curves, we see the amplified noise of the device is greater than SA noise floor. Comparing the NMOS biased and TIA noise floor curves, we know that the TIA noise floor is less than the amplified device noise. 30 Figure 2.19 Power spectral density of output power with/without TIA noise effect. Comparing PSD of output power with or without TIA noise effect in Figure 2.19, we can see the TIA noise has significant effect at higher frequencies. So the TIA noise power should be removed from the total output power to get accurate drain current noise power. Before measuring the drain noise characteristics, the I-V curve of this TIA shown in Figure 2.20 is measured when the DUT is turned off. 31 Figure 2.20 Measured I-V curve of TIA with DUT biased off Figure 2.20 shows this TIA turns on at VCC = 1.5V and the Icc-Vcc behaves like a resistor at VCC > 2.5V. A Vcc = 4.25 V is used in this work. From equation (2.8) totalZ can be calculated by mintotal gV V 1Z out? . (2.14) Figure 2.21 A typical totalZ vs frequency. 32 Figure 2.21 shows the total gain versus frequency at VGS = 0.51V and VDS = 1V. Gain is constant below 2 GHz, drops to half values at 3 GHz, indicating a 3 dB bandwidth of 2.6 GHz. The noiseness in the lower frequency range normally comes from S-parameters. From total gain and measured TIA and buffer noise voltage tiaoV, , PSD of the TIA and buffer current noise BTIA,i , tiaS , is obtained as 2 2,2 ,iS to ta ltiaoBT IAtia ZV?? (2.15) For VGS=0.55V, VDS=1V, tiaS is shown in Figure 2.22. tiaS is 6? 10-23 A2/Hz for frequency lower than 1 GHz, and it increases to 2.5? 10-22 A2/Hz from 1 GHz to 3 GHz. Comparing Figure 2.22 and Figure 2.16, we can see that Sid is smaller than tiaS when frequency is higher than 2 GHz, which indicates the noise from TIA is not negligible at high frequencies in drain current noise measurement. Figure 2.22 A typical tiaS and Sid vs frequency. 33 Chapter 3 Noise Measurement and Data Analysis After boosting the noise though TIA, the output power of the noise can be measured by spectrum analyzer (SA). Drain current noise is extracted from the measured and calculated NMOS + TIA system parameters data. Noise factor ? is calculated from measured Sid in the white noise range of the measured drain current noise spectrum. 3.1 Noise Measurement Figure 3.1a and Figure 3.1b show the noise measurement equipment setup used in our lab. On the left is a HP 4155 semiconductor analyzer, which we use to characterize DC I-V, transconductance gm, output resistance Rout, as well as to supply power to both DUT terminals and VCC of TIA. The metal box in front of the probe station adapts BNC connectors of the cables from the DC biasing probes to the special SMU cable connectors. The output power of amplified drain current noise is measured using a PSA series spectral analyzer, shown on the right. At the TIA output, a bias tee is used between the TIA output and SA to block DC and pass AC. As described in Appendix A and Appendix B, the measurement system needs to change during DC measurement and S-parameters measurement. 34 Figure 3.1a Drain current noise measurement system setup. Figure 3.1b Drain current noise measurement system diagram. 35 Figure 3.2 Block diagram of operation theory for spectrum analyzer [29]. Figure 3.2 shows block diagram of a spectrum analyzer. The input signal to be analyzed and the synchronous local oscillator signal which is generated by gate controlled scan generator are combined at the mixer. After passing the IF resolution bandwidth (RBW) filter, the filtered IF output goes through a log amplifier and is detected by an envelop detector. Then the signal passes through a video bandwidth filter which is used to decrease peal-to-peak variations of noise to a peak/sample detector. Then the signal passes through a video bandwidth filter which is used to decreasing the peal-to-peak variations of noise to a peak/sample detector. The display logic modify the horizontal (frequency) axes. The peak/sample detector detects and passes the signal to analog-digital converter to transfer the analog signal into digital signal to allow the spectrum analyzer to display the signal. Considering the operation theory of SA [27-30], the following SA setting are used: Frequency range: 100 MHz ~ 3 GHz, Resolution Bandwidth (RBW on instrument panel) = 300 KHz, Video Bandwidth (VBW on instrument panel) = 3 KHz, Video Bandwidth/Resolution 36 Bandwidth = 0.01, Reference level = -60 dBm and Attenuation = 0dB. The default value of RBW is 3 MHz on SA we used. After changing it to 300 KHz, we compare these two measured results by overlay them in one figure and the figure shows taht these two curve are totally coincident. It demonstrates that both 3 MHz and 300 KHz work well for RBW setting in our work. The ratio of VBW/RBW is usually set to 0.1 or 0.01, which is small enough for accurate measurement result as described in [28]. Figure 3.3 Die photo with right AC probe and DC probes on pads. Figure 3.3 shows a photo of the die and probe configuration during noise power measurement. A Cascade Infinity GSG probe contacts the GSG pads at the TIA output, and connects the output to SA input. The Eye-Pass DC probe, shown in Figure 3.4, is used to contact all the DC pads and supply power to TIA through the VCC pad, to DUT gate through the VGS pad, and to DUT drain through the VDD pad. The capacitance Cp in Figure 3.1b is 10 nF as shown in Figure 3.4 to keep AC ground from these pads. The ground tip showed in Figure 3.4 is contacted to our designed ground pad one to one. 37 Figure 3.4 Configuration of Eye-Pass DC probes (Cascade Microtech Company). 3.2 Experimental Results Analysis The drain current spectral density and noise factor from all the bias set we chose are shown in Figure 3.5. Figure 3.6 and Figure 3.8 demonstrate the voltage dependence of ? and Sid respectively. 38 Figure 3.5 Drain current noise spectral density Sid and noise factor ? versus VDS. Figure 3.6 Noise factor ? with different biases. 39 Figure 3.7 Extracted noise factor ? vs VGS in 0.18 ?m technology [14]. Same as plots in Figure 3.7, we can see that ? in Figure 3.6 decreases with increasing VGS at fixed VDS, and it increases with increasing VDS at fixed VGS. 40 Figure 3.8 Drain current noise PSD Sid vs Vgs. Figure 3.9 Drain current noise PSD Sid vs Vgs in 0.18 ?m technology [6]. 41 Same as plots in Figure 3.9, Sid in Figure 3.8 increases with raising VGS at fixed VDS and it also increases with increasing VDS at fixed VGS. Figure 3.10 shows that Sid increases with IDS at fixed VDS. This phenomenon is also shown in Figure 3.11 [31]. Figure 3.10 Drain current noise PSD Sid vs drain current IDS. Figure 3.11 Drain current noise PSD Sid vs drain current IDS in 90 nm technology [31]. 42 As mentioned in Chapter 2, TIA is not noiseless and its noise power should be removed from the total measured noise power. Figure 2.15 shows us the noise floor of this SiGe TIA in green curve is only 3 or 4 dBm/Hz lower than the amplified noise power density in red curve when frequency is higher than 1.5 GHz. In other words, the noise of TIA is not negligible at higher frequency. To remove the TIA noise effect from measured drain current noise, we subtract TIA noise power from the total measured output power. And the result is shown in Figure 3.12. Figure 3.12 Comparison of output noise power From Figure 3.12 we can see the measured output power without the effect of TIA which shown in red curve is lower than the total output power which contains the power of TIA noise. This phenomenon is obvious when frequency is higher than 1.5 GHz. 43 After subtracting the TIA noise power from the total measured power, the derived drain current noise PSD and noise factor are shown in Figure 3.13 and Figure 3.14. Figure 3.13 Comparison of Sid Figure 3.14 Comparing of ? 44 Actually, the TIA noise floor which shown in Figure 2.15 is not exactly the same as equivalent TIA noise when DUT is biased. Therefore, there is an approximation. A better way to remove the TIA noise effect from measured drain current noise is need to do as the future work. Moreover, the error bar of noise measurement on SA is approximately 1 dBm, which also affects the final experimental result. Figure 3.15 Sid vs frequency at fixed VGS. 45 Figure 3.16 Sid vs frequency at fixed VDS. Figure 3.15 and Figure 3.16 show Sid versus frequency at fixed VGS or fixed VDS. For both 1/f noise component and thermal noise component, Sid increases with VDS at fixed VGS and Sid increases with VGS at fixed VDS. The corner frequencies only change a little at these bias. 46 Appendix A On-Wafer DC Measurements A.1 Introduction As mentioned in Chapter 1, to derive the drain current noise spectral density and noise factor, we need to measure transconductance (gm) and drain-source conductance (gds). Hence, DC I-V measurements need to be done to acquire these parameters. Besides, we can see the mathematical relationship between the voltages and current at each terminal and know the operating limits of the transistor. Figure A.1 DC measurement system setup 47 This measurement system was controlled by ICCAP software through GPIB-USB connected to probe station and HP4155 semiconductor parameter analyzer which was used to supply DC power through bias box. Figure A.2 Die photo with DUT and TIA on chip One test die under microscope is shown in Figure A.2. The red circle contains the DUT and TIA in circuit. The blank circle contains the standalone NMOS which is designed to be the same structure as the NMOS in the circuit, and the size of the NMOS is the same as DUT in circuit. The reasons putting a standalone NMOS here are using the space on wafer efficiently and see some basic DC and RF characteristics. 48 A.2 DC Measurements for Standalone MOSFET To know to basic characteristics of the transistor, DC measurements for standalone MOSFET was carried out at first. As shown in Figure A.3, two Ground-Signal-Ground AC probes was used to test the standalone NMOS. Figure A.3 Die photo with AC (G-S-G) probes on standalone NMOS pads A.2.1 IDS-VDS and IDS-VGS Measurements Figure A.4 Measured IDS-VDS for standalone NMOS 49 Considering the output characteristic of the transistor shown in Figure A.4, most of the drain biases we chosen, such as 0.46V, 0.73V and1V, are in the saturation regions. To compare the noise characteristics in saturation region which are interested in most cases, we also chose 0.18V in triode region as the drain bias. Figure A.5 Measured IDS-VGS for standalone NMOS Considering the transfer characteristic of the transistor shown in Figure A.5, most of the gate biases we chosen, such as 0.55V, 0.58V, 0.61V and 0.64V, are in the moderate and strong inversion regions. A.2.2 gds and gm In A.2.1, we got the measurement result for IDS-VDS and IDS-VGS. To acquire gds and gm which are used for noise spectral density and noise factor extraction, equation (1.2) and (1.6) are used and the calculation results are shown in Figure A.6 and Figure A.7 separately. 50 Figure A.6 Gate-source conductance gds vs VDS Figure A.7 Transconductance gm vs VGS 51 Also, we can check the output resistance of the transistor Ro from gds, which is shown in Figure A.8. The resistances are several hundreds of Ohms for these bias sets, which require the equivalent of input impedance of TIA part be less to allow drain current noise current flow into TIA. Figure A.8 Output resistance of the NMOS vs VDS A.3 DC measurements for Integrated DUT in Circuit A.3.1 Measurement Equivalent Circuit DC measurements for integrated MOSFET with TIA in circuit are more complex than DC measurements for standalone MOSFET since IDS-VDS and IDS-VGS measurements cannot be measured directly. 52 Figure A.9 DUT in circuit DC measurement circuit Figure A.10 Die photo with one AC probe and DC probes on pads for DC measurement 53 Figure A.9 shows the equivalent circuit for DUT DC measurement. VDD and IN provide DC voltage for drain and gate separately. VDS probe is used to monitor drain voltage by adding a tiny current source iDS at VDS pad, such as 10 nA. With this small current, the voltage drop can be ignored so that the tested voltage shown on VDS is very close to drain voltage. Also, because there is a 2 K? resistor between VDD pad and drain terminal, VDD is not the voltage value we add to drain. In case of damaging the DUT, we have to estimate the VDD using DDDSGS RIV ???DDV (A.1) where VGS is the operating voltage added to drain terminal, RDD = 2 K?, measured current IDD from VDD pad is the drain current IDS which should be similar value in standalone NMOS DC measurement result. Compared with Figure A.9, Figure A.10 tells how the real probe pads matched the pins in the electrical circuit. A.3.2 IDS-VDS measurement and gm Extraction Because IDS-VDS for the DUT cannot be measured directly, we need to do IDD-VDD and VDS-VDD measurement first, and then show the data using IDD-VDS axes. Attention is IDD = IDS as we talked before. 54 Figure A.11 Measured IDS-VDS for integrated DUT in circuit Figure A.11 shows the output characteristic of the integrated DUT in circuit. To compare with the DC measurement result for standalone NMOS and integrated DUT in circuit, the IDS- VDS plots from Figure A.4 and Figure A.11 are overlay by Figure A.12. Figure A.12 Comparing IDS-VDS for both integrated DUT in circuit and standalone NMOS 55 For the manufacture reason, there are slight differences between the DC characteristics of standalone NMOS transistor and the integrated DUT in circuit. The drain current difference increases with raising gate voltage or drain voltage. Similar to standalone NMOS DC measurement, using equation (1.2), gds could be derived from IDS-VDS. And gds of this DUT in circuit is shown in Figure A.13 Figure A.13 gds of integrated DUT in circuit To acquire gm of integrated DUT, there are two methods. The first one is get gm from gm ?VGS measurement data for standalone NMOS. The second method shown in Figure A.14 is chosen very close to values of VGS, using the difference of the nearby IDS values as SDI? and using the difference of the nearby VGS as SGV? . In this method, gm is given by 13 13mg GSGS DSDS VV II ??? (A.2) 56 Figure A.14 IDS-VDS for nearby VGS Keep the Figure A.12 in mind; it is more appropriate to use the second method to derive gm for higher biases set. 57 Appendix B On-Wafer S-Parameters Measurement B.1 Introduction Scattering parameters (S-parameters) are power wave descriptors that used to define the input-output relations of a network, which is a two port network in our work, in terms of incident and reflected power waves. Figure B.1 Convention for defining S-parameters for two-port networks Normalized incident power wave an and reflected power wave bn are given by )( 2 1 nonon IZVZa ?? (B.1a) )( 2 1 nonon IZVZb ?? (B.1b) where n is port number 1or 2, Zo = 50 ? is the characteristic impedance of the connecting lines on the input or output side of the network. Vn and In are input or output voltage and current. And based on Figure B.1, S-parameters are defined by ]][[][ nn aSb ? (B.2) 58 where S11 is the ratio of reflected to incident power wave at port 1; S22 is the ratio of reflected to incident power wave at port 2; S12 is the ratio of transmitted power wave at port 1 to incident power wave at power 2; S21 is the ratio of transmitted power wave at port 2 to incident power wave at power 1. By matching on the input or output side, S-parameters can be determined. Figure B.2 shows the simple equivalent circuit of S11 and S21 measurement, where the line impedance Zo is matched through a corresponding load impedance ZL = Zo for a2 = 0 at the output side. Figure B.2 Measurement of S11 and S21 by matching the line impedance Zo at port 2 through a corresponding load impedance ZL = Zo From Figure B.2, we can calculate S11 by the input reflection coefficient in? : oin oinin ZZ ZZS ?????11 (B.3) On other hand, we can acquire Zin from S11 by changing the form of equation B.3 to 11 1111 SSZZ oin ??? (B.4) Besides, from equations B.1a, equation B.1b and Figure B.2, we can calculate S21 by 11 20 1 221 2|S 2 IZV Vab oa ??? ? (B.5) 59 where V1 = VS1 - ZoI1 and V2 is the output voltage Vout. Hence, we can reform equation to 1 out21 2S sV V? (B.6) Since the voltage at port 2 is directly related the source voltage, available forward voltage gain can be get from S21 in dB unit by ||log20 21, SG avv ? (B.7) Figure B.2 also represent that input voltage can be acquired by ino insin ZZ ZVV ?? (B.8) Similar to measurement of S11 and S21 the measurement circuit (Figure B.3) of S22 and S12 can be considered as a reflection of Figure B.2 by matching the line impedance Zo through a corresponding input impedance ZG = Zo Figure B.3 Measurement of S22 and S12 by matching the line impedance Zo at port 1 through a corresponding input impedance ZG = Zo 60 B.2 S-Parameters Measurement System For S-parameters measurement system setup shown in Figure B.4, GPIB-USB was used to connect Agilent PNA E8464B network analyzer, HP4155 and a computer which offered ICCAP software. After setting up the measure/simulation and instrument option sections in ICCAP, HP4155 and PNA can be controlled automatically. HP4155 which works as DC power supply and PNA which works as AC power supply are connected to Probes through bias T. Power meter is connected to PNA by GPIB-USB to help with the power calibration portion during S-parameters measurement. Figure B.4 S-parameters measurement system 61 Figure B.5 Simplified block diagram of the PNA network analyzer system As shown in Figure B.5, a phase-locked incident signal which includes both reference signal and test signal is generated from the synthesized source. Through the signal separation portion of the PNA, the reference signal was received by the receiver portion and the test signal is applied to the device under test (DUT) through port 1 and port 2. In this work, port 1 is selected as the source port as shown in Figure B.6. In the test system, random and systematic measurement errors are involved in the S- parameters measurement. Although some random errors, such as thermal drift, cannot be removed systematically, the systematic error of the PNA network analyzer and the power loss of the cables can. In our work, the calibration standard technique short-open-load-through (SOLT) which uses testing and calculating short, load and through standard parameters to remove errors is used with Cascade impedance standard substrate ISS 101-190B which fit our AC probes and shown in Figure B.6. 62 Figure B.6 Layout of Cascade impedance standard substrate 101-190B (Cascade Microtech Inc.) B.3 S-Parameters Measurement for Standalone MOSFET Similar to DC measurement, S-parameters measurement for standalone MOSFET was carried out first to check the setup of measurement system, the measurement method and RF performance of the transistor. The S-parameters measurement circuit is shown in Figure B.7. The DC power from HP4155 and AC power from PNA are combined in bias Ts and the output signals go into the NMOS through AC pads IN and OUT. 63 Figure B.7 Circuit of the S-parameter measurement system for standalone NMOS Take biases set VGS = 0.55V and VDS = 1V (IDS = 4.5 mA) for example, the measurement results is shown in Figure B.8: Figure B.8 Real and image parts of S-parameters versus frequency for standalone NMOS Changing these data into dB unit by SdB = 20 log |S|, we can see Figure B.9 64 Figure B.9 S-parameters versus frequency for standalone NMOS in dB unit As we mentioned in section B.1, S21 and S12 are the forward and reverse voltage gain separately. In Figure B.9, we can see the forward gain S21 is between 13dB to 14dB and it begins to drop a little at higher frequency. Reverse gain S12 is between -40dB to -20dB which is very small compared to other S-parameters. And it increases with rising frequency. Moreover, S11 and S22 is around 0 dB. These S-parameters represent that this NMOS transistor work well at the bias set VGS = 0.55V and VDS = 1V. Besides S-parameters, the admittances Y-parameters are also common multi-port network parameters to check RF characteristics. Y-parameters of two-port network are usually defined by ][V][]i[ nn Y? (B.9) We can extract Y-parameters from S-parameters as described in by 65 21122211 21122211 o11 SSS1S1 SSS1S-1Z1Y ????? ???? ?? ??? (B.10a) 21122211 12 o21 SSS1S1 2S-Z1Y ????? ??? (B.10b) 21122211 21 o12 SSS1S1 2S-Z1Y ????? ??? (B.10c) 21122211 21122211 o22 SSS1S1 SSS1S1Z1Y ????? ???? ?? ???? (B.10d) And the calculation results is shown in Figure B.10 Figure B.10 Y-parameters versus frequency for standalone NMOS 66 B.4 S-Parameters Measurement for integrated DUT and TIA System The electrical circuit for S-parameters measurement for DUT and TIA is shown in Figure B.11. Figure B.11 S-parameters measurement system circuit In this measurement system, the DUT and TIA are considered as a whole system. HP4155 is used to provide DC power for NMOS and TIA through RF pad IN, DC pads VDD and VCC. In this measurement, DC pads, RF pads IN and OUT all are connected to the probes like shown in Figure B.12. 67 Figure B.12 Die photo with two AC probe and DC probes on pads for S-parameters measurement One attention for S-parameters measurement is choosing the power value on PNA. When frequency = 1GHz, Vds = 1 V and Vgs = 0.51 V, as shown in Figure B.13, the source power chosen for our work is -60 dBm which is small enough to neglect its effect on DC characteristics and also not too small to cause the RF characteristic distortion for our project. Figure B.13 Power sweep for S-parameters measurement 68 For an instance, the tested S-parameters for the DUT and TIA system at bias set VGS = 0.51V, VDS = 0.18V is shown in Figure B.13 Figure B.14 S-parameters versus frequency for NMOS + TIA In Figure B.14, S21 of the NMOS + TIA system keeps staying at 42 dB from 100 MHz to 1.50 GHz at 42 dB, and after 1.50 GHz, it drops from 28 dB to 23 dB. The reverse voltage gain S12 is between -60 dB to -40 dB. S11 keeps staying at 0 dB at lower frequency and it begins to drops approximately after 1.50 GHz. At 3 GHz, it drops to -3 dB. S22 decreases from -1 dB to -4 dB from 400 MHz to 3 GHz. Besides, we can get equivalent input impedance of this NMOS + TIA system Zin from S- parameters. 69 Figure B.15 Equivalent input impedance versus frequency for NMOS + TIA Figure B.15 shows us that the real part of the equivalent of the input impedance seems like a resistor which values R = 36 ? and is almost not changed with frequency after 500 MHz. Neglecting the effect of inductance, the image part of the input impedance ?? inZ? is given by Ci?1Z in ?? ?? (B.111) From equation B.13, we can derive equivalent input capacitance C to be ?? inZ1-C ?? ? (B.12) 70 Figure B.16 Equivalent input capacitance versus frequency for NMOS + TIA Equivalent input capacitance shown in Figure B.16 is around 0.8 pF. The equivalent input impedance can be considered as the resistor and capacitor in series. From Figure B.15 and Figure B.16, the magnitude of the equivalent input impedance can be calculated by Ci?1RZ in ?? . (B.13) The magnitude of equivalent input impedance is shown in figure B.17. 71 Figure B.17 Magnitude of equivalent input impedance for NMOS + TIA 72 Bibliography [1] A. van der Ziel, "Noise in Solid-state Devices and Lasers," proceeding of the IEEE, vol. 58, No. 8, August 1970. [2] Y. Nemirovsky, I. Brouk and C. G. Jakobson, "1/f Noise in CMOS Transistors for Analog Applications," IEEE transactions on electron devices, vol. 48, No. 5, May 2001. [3] A. Hati, D. A. Howe, F. L. Walls and D. Walker, "Noise Figure vs. PM Noise Measurements: A Study at Microwave Frequencies," IEEE International Frequency Control Symposium and PDA Exhibition, 2003. [4] Y. Tsividis, Operation and Modeling of The MOS transistor, McGraw-Hill Inc., 1999. [5] J. B. Johnson,"Thermal Agitation of Electricity in Conductors," Phys. Rev. 32, 97?109, 1928. [6] A. Van der Ziel, "Thermal Noise in Field-Effect Transistors", Proc. of the IRE, vol. 56, pp. 1808-1812, August 1962. [7] C. Chen and M. J. Deen,"Channel Noise Modeling of Deep Submicron MOSFETs, "IEEE IEEE transactions on electron devices, vol. 49, No. 8, August 2002. [8] J. C. J. Paasschens, A. J. Scholten, and R. van Langevelde, "Generalizations of the Klaassen- Prins equation for calculating the noise of semiconductor devices", IEEE transactions on electron devices, vol. 52, No. 11, November 2005. [9] A. A. Abidi, "High-Frequency Noise Measurements on FET?s with Small Dimensions," IEEE transactions on electron devices, vol. ED-33, No. 11, November 1986. [10] S. Dronavalli and R. P. Jindal, "CMOS Device Noise Considerations for Terabit Lightwave Systems,? IEEE transactions on electron devices, vol. 53, No. 4, April 2006. [11] C. Chen, "Thermal Noise in Modern CMOS Technology," Solid State Circuits Technologies, InTech, 2010. [12] C.Chen, M.J.Deen, M. Matloubian and Y. Cheng, "Extraction of the induced gate noise, channel thermal noise and their correlation in sub-micron MOSFETs from RF noise measurements," Microelectronic Test Structures, 2001. ICMTS 2001. [13] B. M. Wilamoski and D. Irvin, Fundamentals of Industrial Electronics, Taylor & Francis Group, 2011 73 [14] K. Han, K. Lee and H. Shin, "Thermal Noise Modeling for Short-Channel MOSFET?s," Simulation of Semiconductor Processes and Devices, SISPAD 2003. [15] K. Yan, Noise Measurements, Models and Analysis in GaAs MESFETs Circuit Design, PHD dissertation, Oregon State University, 1996. [16] H. A. Haus et al., "Representation of Noise in Linear Twoports", Proc. IRE, vol. 48, no. 1, pp. 69?74, Jan. 1960. [17] H. Rothe and W. Dahlke, ?Theory of noisy fourpoles,? Proceedings of the IRE, Vol. 44, , pp. 811?818, June 1956. [18] L., R. Q., "The determination of device noise parameters". Proc. of the IEEE, Vol. 57, No. 8, pp. 1461-1462, Aug. 1969, [19] M. Mitama, and H. Katoh, "An improved computational method for noise parameter measurement," IEEE Trans. Microwave Theory Tech., Vol. MTT-27, No. 6, pp. 612-615, June 1979. [20] G. I. Vasilescu, G. Alquie, and M. Krim, "Exact computation of two-port noise parameters," Electron. Lett. , pp. 292-293, Vol. 25, No. 4, Feb. 1988. [21] O'Callaghan, J.M. and Mondal, J.P,"A vector approach for noise parameter fitting and selection of source admittances",IEEE Trans. Microwave Theory Tech., Vol. 28, No. 8, pp. 1376-1382. Aug.1991. [22] C. H. Chen, et al., "Novel noise parameter determination for on-wafer microwave noise measurements". IEEE Trans. Instrum. Meas., Vol. 57, Issue 11, pp. 2462-2471, May 2007. [23] L. F. Tiemeijer, R. Havens, R. Kort, and A. J. Scholten, "Improved Y-factor method for wide-band on-wafer noise parameter measurements," IEEE Trans. Microw. Theory Tech., vol. 53, no. 9, pp. 2917?2925, Sep. 2005. [24] S. Tedja, H.H. Wiliams, J. van der Spiegel, F.M. Newcomer and R. Van Berg. "Noise Spectral Density Measurements of a Radiation Hardened CMOS Process in the Weak and Moderate Inversion," IEEE Transactions on Nuclear Science. Vol. 39(4). pp. 804-808, 1992. [25] S. Tedja, "Analytical and Experimental Studies of Thermal Noise in MOSFET's", IEEE transactions on electron devices, vol. 41, No. I I , Novenber 1994. [26] J. Cressler and G. Niu, Silicon-Germanium Heterojunction Bipolar Transistors. Boston, MA/USA: Artech House, INC., 2002. [27] Making RF Power and Frequency Measurements with Spectrum and Signal Analyzers, Agilent Technologies, Inc. 2012 74 [28] Agilent Technologies E8362B, E8363B and E8364B Option H85 User?s and Service Guide Supplement, Agilent Technologies, Inc. 2004, 2005, 2006. [29] Measurement Guide and Programing Examples PSA and ESA Series Spectrum Analyzers, Agilent Technologies, Inc. 2004 [30] R.Ludwig, P.Bretchko, RF Circuit Design: Theory and Applications, Prentice Hall, New York, 2000 [31] X. Wei, G. Niu, Y. Li, et al., ?Experimental characterization and simulation of RF intermodulation linearity in a 90 nm RF CMOS technology?, in Dig. of IEEE Radio Frequency Integrated Circuits Symp., pp. 251-254, 2008.