Chapter 3 Experiments
3.3 Measurements
3.3.2 RF Measurements
The RF characteristics were measured to evaluate the effects of using copper airbridges on the device performance.
a.) S-parameters measurement
The s-parameters were measured by voltage-traveling waves with a vector network analyzer (HP8510C).
The scattering parameters (S-parameters) are four complex numbers,
S11, S12, S21 and S22, which represent the relationship between incident and reflected voltage waves (having amplitude and phase) at the input and output ports of a FET.
The S-parameter measurements can reveal gain, loss, reflection coefficient, etc. of a device. In order to obtain scattering parameters (S), we need to measure the incident and reflected waves at ports 1 and 2 of the two-port network. From Figure 3-6, the relationship between a1, a2, b1, and
b2, can be expressed as: waves to the incident waves with proper terminations.
1 0
S11 is the input reflection coefficient with the output matched.
2 0
S12 is the reverse transmission coefficient with the input matched. It is also a measure of the isolation of the device.
1 0
S21 is the forward transmission coefficient with the output matched; it also measures the voltage gain of the device.
2 22 = b
S (5)
S22 is the output reflection coefficient with the input matched.
S-parameters do not require short circuits or open circuits during the measurement. It only requires that the device be terminated in matched impedance equivalent to the characteristic impedance of the transmission line.
b.) Equivalent circuit model
An equivalent circuit model is useful for circuit design. Another important function of S-parameters is to allow the calculation of equivalent circuit parameters of the device. Figure 3-7 illustrates the equivalent circuit components in a PHEMT structure based on a physical view point. The small signal equivalent circuit model is shown in Figure 3-8, which includes two parts: intrinsic and extrinsic. The elements in the box labeled “intrinsic”
represent the fundamental elements responsible for the inherent nature of the PHEMT device. Other resistors and inductances, known as extrinsic elements, generally represent undesirable parasitic elements that can be suppressed by good design and process, but would never be eliminated. The small–signal model is required to analyse the measured S-parameter at microwave frequencies. Computer modeling programs are used to determine the circuit
parameters that will generate measured scattering parameters. The H, Y, or Z parameters can be derived from the S-parameters measured. Relationships exist between S-parameters and other two-port network parameters. For Y-parameters, the following relationships are used:
( )( )
Therefore, the circuit modeling will be established. Modeling process is as follows.a). Intrinsic model value
1. Measurement of device S-parameters 2. Translate S-parameters into Z-parameters b). Parasitic impedance
2. Translate S-parameters into Z-parameters 3. Deduct parasitic impedance
The study of small signal equivalent circuit models of copper-airbridged and gold-airbridged PHEMTs was made to investigate the RF performance difference. The source resistance (Rs) with copper airbridges and gold airbridges can be derived from the small signal equivalent circuit models with fitting the measured S-parameters.
c.) Noise Figure
Three sources of noise resulted from circuits are thermal noise, shot noise and flicker noise. Thermal noise is caused by the thermal agitation of the carriers in the ohmic resistance. Shot noise is a current dependent effect caused by fluctuations in the electron and hole currents due to bias conditions.
The flicker noise, which manifests itself at low frequencies (usually 0.01 kHz – 100 kHz) with the 1/f spectral density dependence, is caused by carrier density fluctuation and mobility fluctuation.
The intrinsic noise sources in a GaAs FET are the thermal-generated channel noise and the induced noise at the gate. For pHEMT, at high frequencies (＞ 1 GHz), the flicker noise(1/f noise) sources are unimportant, and the transistor channel noise is mainly due to thermal noise of the carriers if
the device is biased into the Ohmic region of the DC characteristic. If the device is biased at the saturated region, shot-like noise also contributes. The extrinsic noise sources are associated with the Rg and Rs resistances and the gate bonding pad resistance.
Noise figure is also a function of the operating bias condition. In general, the optimum impedance match for lowest noise figure will not be the optimum impedance match for maximum gain. Hence, noise figure measurements are often accompanied by associated gain, the gain achieved at that impedance match.
The noise performance of a FET may be quantified by the noise figure,
NF, which is a function of frequency, FET bias voltages, and impedance matching. Noise figure is an important characteristic of an amplifier, especially for one intended to amplify weak signals, as in the receiver applications. These amplifiers are commonly called LNAs. The lower the noise figure is; the weaker the signal that can be detected after amplification is.
Noise figure reflects the noise added to the signal by the imperfect amplifier, and is defined as the signal-to-noise ratio (S/N) of the input signal divided by the signal-to-noise ratio of the output signal,
It is expressed in dB:
⎥⎥
(Base 10 logarithms). Where,
Si：Input signal power, So：Output signal power, Ni：Input noise power, No：Output noise power, SNR：Signal to noise power ratio
The optimum noise figure, NFo, occurs when the biases and impedance matching are optimized to obtain the best noise figure for the device at a given frequency.
In order to achieve the lower noise performance, a low noise FET should be with a short gate(0.5μm or less) to minimize Cgs, high channel doping to increase gm and decrease Rs, and a short source-to-gate spacing to decrease Rs.
CHAPTER 4
Results and Discussions
From the above study, we know that barrier layer should have inherent characteristics of the high melting point and lower solubility in copper even though at high temperatures. But a trade-off between the barrier property and the adhesion performance with Cu will be made. Barrier material may reveal excellent barrier property but poor adhesion if the barrier does not react with Cu to form compounds.
An ideal barrier material should be with good barrier property as well as good adhesion with Cu. But it is hardly to achieve it. New multilayer system, Ti/WNx/Ti/Cu, was used to solve this problem for getting good film adhesion and good diffusion barrier properties.
In this chapter, the results of the copper airbridges are described. DC and RF performances are also presented. Furthermore, a small signal equivalent circuit derived from measured S-parameters is established in order to compare the performance of the copper-airbridged and gold-airbridged PHEMTs.
4.1 Thin Metal Etching
In order to obtain fine surface with minimum roughness for RF signal transferring, we performed wraparound etch experiments on the monitor wafer
with copper airbridges of the scheme of Ti /WNx /Ti /Cu for finding the optimal ratio of chemical compositions of the etching solution.
The thin copper metal was etched by diluted hydrogen peroxide-sulfuric acid solution with various ratios of constituents. The experiments were performed by using the solution of H2SO4: H2O2: H2O with the ratios of 5:3:100, 5:6:100, 5:9:100, 10:6:100, 15:6:100, 1:1:20, 1:1:50, and 1:1:100, respectively. The experiment data were measured by KLA-Tencor P10 and were shown in Table 4-1 and Figure 4-1a. From Figure 4-1a, the surface roughness of the thick copper interconnects was slightly changed when etched in the solution with the ratio of 1:1:100 (5ml: 5ml: 500ml). When etched in the solution with the ratio of 1:1:50 (5ml: 5ml: 250ml) for 12 seconds, the surface roughness of the plated thick copper was slightly reduced. When etched in the solution with the ratio of 5:6:100 (10ml: 12ml: 200ml), the surface of the thick copper was very shining.
The roughness value reduced a lot and was etched quickly and the film just last for 6 seconds. But when etched in the solution with the ratio of 1:1:20 (5ml: 5ml:
100ml), the surface roughness was increased and was etched off rapidly in 3 seconds. The ratio of 1:1:20 is close to the ratio of 5:6:100, but the results are so different. So we need to check more ratios of the solution constituents.
Next, the experiments by the ratios of 1:1:20, 1:1:30, 1:1:50, 1:1:100,
5:4:100, 5:5:100, 5:6:100, 5:7:100, and 5:8:100 were performed. And the experiments data were shown in Table 4-2 and Figure 4-1b. From Figure 4-1b, among the first group of 1:1:20, 1:1:30, 1:1:50, and 1:1:100 ratios, the solution ratio of 1:1:100 resulted in the reduction of the surface roughness. From another group of 5:4:100, 5:5:100, 5:6:100, 5:7:100, and 5:8:100 ratios, the solution ratio of 5:6:100 resulted in the reduction of the surface roughness. But the etching time of the ratio of 1:1:100 was 30 seconds, which is longer than the 6 seconds etching time of the ratio of 5:6:100. To avoid dipping the sample in the etching solution too long, which would result in some uncontrollable failure, we decided to use the optimal ratio of H2SO4 to H2O2 and H2O with 5:6:100. The solution ratio of 5:6:100 would give rise to the shining copper surface and took a short etching time.
After the copper thin metal etching test, we had run a series of experiments to optimize the etching parameters of the various thin metal layers. The optimized etching recipes of the thin metal layers were shown in Figure 4-2. The optimal processes had been described in detail in Chapter 3.
The thin metal was also tested with using the solution mixed from NH4OH/H2O2/H2O of the ratio 1:1:5 to etch all the layers. Cu was etched severely and more rapidly than WNx. And WNx protruded at the bridge edges (Figure 4-3).
It was a non-selective etchant. However, the solution mixed from H2SO4/ H2O2/ H2O was a selective etchant. The profile of Cu airbridge used of H2SO4/ H2O2/ H2O etching revealed better than that used of NH4OH/H2O2/H2O solution (Figure 4-4).
4.2 AES Depth Profile Analysis
Blanket samples with Ti/WNx/Ti/Cu multilayer were used to study the material property of the thin metal systems. The Au layer was deposited on GaAs blanket wafer and followed by sputtered 300 Å Ti, which was used as the adhesion layer between WNx and Au-contacts, and 400 Å WNx, was used as the diffusion barrier for copper. Then Ti/Cu layers, which used as the adhesion and the conducting seed layer for the subsequent copper electroplating, were applied by evaporation. The thicknesses of Ti and Cu layer were 300 Å and 1000 Å respectively.
The wafer was subsequently split up to four pieces. One was as deposited and without annealing; the others were separately annealed at 300°C, 400°C and 500°C for 30 minutes in the oven. Then Auger Electron Spectroscopy (AES) with model of VG Scientific Microlab 350 were used to analyze the depth profiles of all these four samples.
The AES depth profiles of the Ti/WNx/Ti/Cu multilayer scheme, which were annealed at various temperatures, were shown in Figure 4-5a~c. From the results of these profiles, the diffusion barrier WNx was thermally stable even up to 300°C thermal annealing for 30 minutes. When the annealing temperature was higher than 400°C, the profile of WNx remained changed and the copper began to diffuse into the WNx layer.
4.3 X-ray Diffraction Patterns (XRD)
The XRD data were measured by detector scan with 5 degree fixed axis setting, using the Siemens Diffraktometer D5000 system. The XRD patterns of Ti/WNx/Ti/Cu scheme after thermal annealing at different temperatures for 30 minutes are shown in Figure 4-6. Intermetallic compounds of Au and Cu were formed after the annealing at 500°C. Compared to the AES depth profiles, these intermetallic compounds resulted from that copper atoms diffused through the WNx layer and formed several kinds of intermetallic compound with gold atoms.
(Figure 4-7 is Au-Cu phase diagram.) The resistances of these intermetallic compounds are higher than those of copper and gold, which impacts the RF performance of the LN-PHEMTs. Although there were no obvious intermetallic compounds formed after 400°C annealing from the XRD patterns shown, the
copper diffusion through the WNx layer were observed in the AES profile. The AES profiles indicate that the barrier property of WNx was weak after annealing at such a high temperature. But from the results of AES and XRD, WNx was still a good diffusion barrier between copper and gold even after 300 °C annealing for 30 minutes.
4.4 Adhesion Layer
A Ti adhesion layer was added to solve the copper-airbridged peeling problem. Titanium was extensively used as the adhesion layer in the Si IC industry. The pull-strength of the BLM test systems with Ti adhesive metal was superior to those with Cr [15]. Ti was easily wetting on gold and also conventionally used in the fabrication of gold airbridges. Mainly, Ti can be selectively etched by diluted hydrogen fluoride solution in the thin metal removal process. The deposition and etch processes of Ti are compatible with the conventional processes of GaAs device fabrications.
Airbridge peeling off were not observed again when processing in a tank with ultrasonic vibration for removing the first via photoresist. Also, the diffusion barrier layers with Ti adhesion layers stay quite stable in the scotch tape peeling test. As we sampled 72 dies randomly to measure the electrical performance,
there were only 9 dies failed. The yield of Ti/WNx/Ti/Cu structure was about 85.5% and was better than those of the Au/WNx/Cu scheme, which the yield was just 62.5%.
Figure 4-8 and Figure 4-9 show the uniformities of the DC characteristics of the GaAs PHEMTs samples, which were fabricated with different thin metal systems. The average transconductance of devices using Ti/WNx/Ti/Cu as thin metals was greater than those of WNx/Cu. The standard deviations of Gm and Vp of GaAs PHEMTs fabricated with WNx/Cu were 97 mS/mm and 0.22 V. The standard deviations of Gm and Vp of GaAs PHEMTs fabricated with Ti/WNx/Ti/Cu were 33 mS/mm and 0.14 V. GaAs PHEMTs with Ti/WNx/Ti/Cu scheme showed much better uniformity of Gm and Vp than those with WNx/Cu thin metal layers.
The standard deviations of Gm and Vp of GaAs PHEMTs fabricated with gold airbridges in the same wafer were 45 mS/mm and 0.11 V. So the uniformity of Gm and Vp of GaAs PHEMTs using copper airbridges with Ti/WNx/Ti/Cu multilayer system was comparable to those fabricated with using gold airbridges.
When using Ti/WNx/Ti/Cu as the thin metal systems, the peelings of the plated metal no longer occurred on the LN-PHMTs.
4.5 DC Characteristics
The gate width of LN-PHEMT was 40μm × 4, and the gate length was 0.25μm. Low Noise PHEMT with copper airbridges using of Ti/WNx/Ti/Cu as thin metal layers was annealed at 200℃ for 3 hours in the air to test the thermal stability of the diffusion barrier, WNx. The DC and RF characteristics before and after the thermal annealing were measured and compared. The comparison of the drain I-V characteristics on the same copper-metallized LN-PHEMT with Ti/WNx/Ti/Cu scheme was shown in Figure 4-10a and the dependence of the transconductance on the gate bias voltage was shown in Figure 4-10b. The saturated drain current was about 200 mA/mm and the maximum transconductance was almost up to 449 mS/mm when tested at VDS = 1.5 Volts and VGS = -0.05 Volts. These DC characteristics showed little change after thermal annealing. The saturated drain current after thermal annealing was higher than that before about 11 mA/mm at VDS=1.5V. The knee voltage of the drain I-V curve remains almost the same after the thermal annealing.
To compare with the copper-airbridged PHEMTs with only WNx/Cu as thin metal, for which the saturated drain current was about 150 mA/mm and the maximum transconductance was 375 mS/mm when VDS = 1.5 Volts and VGS = 0 Volts (Figure 4-11a, b), the DC characteristics of copper-airbridged with
Ti/WNx/Ti/Cu as thin metal system were better than those with only WNx/Cu.
The DC characteristics of the fabricated gold-airbridged PHEMTs were compared with that of the copper-airbridged devices. The saturated drain current of the gold-airbridged PHEMTs was about 180 mA/mm and the maximum transconductance was 452 mS/mm when VDS = 1.5 Volts and VGS = 0 Volts. Both devices have the same knee voltage of 0.3 Volts. So the additional Ti layers do not have significant impact on the DC performance of the devices.
4.6 RF Characteristics
The RF characteristics were measured to evaluate the effects of using the Cu airbridges on the device performance. The airbridge of the device is responsible for the parasitic resistance and parasitic inductance and exhibits different impedance at high frequencies. If the airbridges were with poor adhesion and insulation on the device, there would be much attenuation during the signal propagation.
4.6.1 S-parameters Measurement
The S-parameters were measured by HP8510C before and after Si3N4
passivation and thermal annealing at 200℃ for 3 hours. The results are shown in Figure 4-12a,b and Figure 4-12c,d. The comparison is shown in Figure 4-13a,b.
Both the s-parameters were measured under a DC supply-voltage of Vds=1.5 V and the bias voltage Vg = -0.2 V. There is a little change of the S-parameters for devices before and after thermal annealing.
Figure 4-14 shows the curves of magnitude of S21, which presents the gain of the GaAs LN-PHEMT. The values of S21 are almost the same before and after thermal annealing shown. The chart of S12 indicates the isolation of the GaAs LN-PHEMTs. As shown in Figure 4-15, the magnitude values of S12 before thermal annealing are little higher than those after thermal annealing about 0.1 dB at high frequency region. The isolation of after thermal annealing maybe would be better than that before thermal annealing. This means the Si3N4 passivation and the thermal annealing treatment do not influence the gate-drain negative feedback.
There would be not much attenuation in the signal propagation after the device was passivated with Si3N4 and with the thermal annealing treatment.
4.6.2 Equivalent Circuit Model
The study of small signal equivalent circuit model is made to investigation the influence of the airbridges on the RF performance of the GaAs LN-PHEMTs.
The equivalent circuit-modeling diagram was shown in Figure 4-16. It includes two parts: intrinsic and extrinsic. The intrinsic part is FET itself, and extrinsic parts
are the other components. The calculated s-parameters of equivalent circuit model are fitted to the measured S-parameters of Au-airbridged, Cu-airbridged with WNX/Cu, and Cu-airbridged devices with Ti/WNx/Ti/Cu within the frequency range of 10~18 GHz after thermal annealing (shown in Figure 4-17 a~d). From the simulations, we get those parameters values as following:
g =130 mS, Cgs=0.247 pF, Cdg=0.056 pF, Rs=3.3 ohms, for Au-airbridged m
device;
g =135 mS, Cgs=0.25 pF, Cdg=0.05 pF, Rs=5.5 ohms, for Cu-airbridged m
device with WNx/Cu;
g =128 mS, Cgs=0.245 pF, Cdg=0.045 pF, Rs=3.5 ohms, for Cu-airbridged m
device with Ti/WNx/Ti/Cu before thermal annealing;
g = 135 mS, Cgs=0.23 pF, Cdg=0.055 pF, Rs=3.5 ohms, for Cu-airbridged m
device with Ti/WNx/Ti/Cu after thermal annealing.
Wheregm, Cgs, and Cdg are intrinsic parameters, Rs is extrinsic parameter.
As we know, the cut off frequency equation can be expressed as:
(
gd)
g ：the transconductance, m
Cgs：the gate-source capacitance, Cgd：the gate-drain capacitance.
Using the simulation data, we can calculate fT by the above equation and get:
T
T
T
T
f =68.3 GHz, for Au-airbridged device;
f =71.62 GHz, for Cu-airbridged device with only WNx/Cu;
f =70.248 GHz, for Cu-airbridged device with Ti/WNx/Ti/Cu before thermal annealing;
f =74.95 GHz, for Cu-airbridged device with Ti/WNx/Ti/Cu after thermal annealing; these values are similar to 70GHz shown in Figure 4-19 and
f =74.95 GHz, for Cu-airbridged device with Ti/WNx/Ti/Cu after thermal annealing; these values are similar to 70GHz shown in Figure 4-19 and