PROSPECTS OF MICROSTRIP WAVEGUIDES IN ALUMINUM AND COPPER METALLIZATION FOR HIGH-FREQUENCY APPLICATIONS

An electrical characterization of the high­ frequency behavior of monolithically integrated microstrip waveguides and a grounded coplanar line is presented. Mi­ crostrip lines fabricated with both aluminum metallization and copper metallization are compared in detail. Scaling laws and future metallization technologies are discussed. The waveguides are characterized by S-parameter measurements in a frequency range from 100 MHz to 80 GHz. After deem­ bedding the characteristic impedance, propagation constant and telegraphers equation transmission parameter are ex­ tracted. Measurement results are compared to simulation re­ sults in detail. Advantages of aluminum and copper metal­ lization are discussed.


INTRODUCTION
The success of the semiconductor industry is attributed to the advances in miniaturization. Due to the continu ous shrinking of the minimum feature size, which is given by lithography limits, the devices are getting faster and faster Compared to 1970 we can fabricate today 100 times smaller structures on the chip, where more than 20,000 times more transistors with faster switching times are integrated. Traditional scaling will probably no longer satisfy perfor mance requirements. While technology scales down to nano dimensions an ever increasing disparity between gate and wiling delay appears. The interconnect opportunities have to be considered for future integration. Nowadays silicon-based monolithic microwave integrated circuits tend to very high frequencies [2]. The designer has to pay attention to interconnect design, which influences the performance of ICs significantly [3]. Connections between circuit core and pads are long on-chip interconnects and in many cases realized as microstrip lines. The interconnect should meet many demands like low loss or to match a certain impedance.
Recent papers have presented several methods for charac terizing transmission lines by the characteristic impedance  S-parameter measurement [6] with a thru-reflect-line (TRL) calibration [5,7]. This paper shows an accurate characterization of mono lithically integrated microstrip waveguides realized in copper and aluminum metallization and a coplanar waveguide up to 80 GHz. Section 2 shows modern metallization processes and the interconnect requirements addressed in the International Technology Roadmap for Semiconductors (ITRS). Scaling scenarios of interconnects are discussed in Section 3. Sec tion 4 shows the differences between aluminum and copper metallization. In Section 5 the micros trip line geometries are described. For accurate measurement results in Section 6 the deembedding procedure is explained. Section 7 discusses the theory of S-parameter based transmission line characteriza tion. In Section 8 the experimental results of the waveguides are presented and compared to simulation results in detail.

MULTILEVEL METALLIZATION
In next and over-next chip generations on-chip communi cation in terms of fast signal transmission within and between functional blocks is mainly determined by the interconnects or wiling systems. The basic functions of wires connecting the millions of transistors are the distribution of analog and digital signals as well as the distribution of power and sup ply voltage. For higher packing density and design flexibility the third dimension is used for interconnection. The resulting network of interconnects is called multilevel metallization, where metal interconnects span several layers isolated by the interlevel dielectric and connected vertically by vias (Fig. I).
According to the interconnect dimension the backend-of-line (BEOL) can be grouped into different types of wires. At local level short interconnects with minimum feature size are used to connect transistors. Intermediate wire dimensions are used for longer distances and wide interconnects with lengths up to 4 mm [8] are located at the upper or global level.

SCALING OF INTERCONNECTS
Global interconnects with large cross-sectional areas are mainly used, where high current densities and/or low resis tances are needed. Due to the different scaling behavior of local and global interconnects the different levels of inter connects must be considered separately. Based on funda mental scaling laws assuming the same scaling factor s < 1 for all geometry dimensions the RC-delay of local wires re mains constant for minimum feature sizes larger than 100 nm ( Fig.2(a)).  For smaller wires, called nano-interconnects, the electrical size effect leading to increase in resistivity has to be taken into account [10]. The scaling scenario for global wires is slightly different. Due to the fact that global wires do not scale in length the wiring RC-delay of global wires scales withs t " .

'Z! 'e Cw i r e ~ s-2
(2) R 'L This is the dominant contribution to the overall signal de lay time for long interconnections ( Fig. 2(b)) [9]. Therefore advances in all disciplines of microelectronics are required to overcome this interconnect bottleneck and to guarantee highest chip performance for upcoming technologies. From the architecture perspective an aggressive hierarchical wiring system at the expense of additional wiling levels will be im plemented in the BEOL in future chips [II]. This trend is also addressed in the International Technology Roadmap for Semiconductors (Fig. I) [I]. The decreasing metal pitch and the increasing number of metal levels with increasing tech nology node are displayed in this roadmap (Tab. I).

Revista da Sociedade Brasileira de Telecornunicacoes
Volume 18, Nurnero 1, Junho de 2003 In order to investigate the high frequency behavior of global interconnects. microstrip lines fabricated in different metallization technologies are electrically characterized and compared. In recent years, aluminum was replaced by cop r er due to its higher conductivity resulting in less power re quired on chips and due to its better electromigration en durance. Conventionally, plasma etching was widely used for aluminum interconnect patterning. In this subtractive pro cess aluminum is deposited as a blanket film on an adhesion layer; titanium nitride is usually used. For lithography reason an anti-reflecting coating (ARC) is deposited onto the alu minum layer followed by the metal etch through the mask. Finally the mask is removed to form the metal lines (Fig. 3).
With the introduction of copper, new integration schemes are necessary because no viable copper etch technology is currently available. To successfully integrate copper as mate rial for interconnect applications the technology is shifting to damascene processing. In this in-laid technique, first trenches and/or vias are etched followed by the deposition of the metal stack including a barrier layer to prevent copper from diffu sion into the dielectric and a seed layer acting as wetting layer for the subsequent electrochemical deposition. The metal ex cess is removed by chemical-mechanical polishing ( Fig. -l), Due to different integration schemes the impact of the dif ferent barrier layers has to be considered. For the case of copper, which is a fast diffuser, barriers are needed at the trench bottom and on both sides of the trench. The barrier thickness will decrease with shrinking feature sizes in order  to keep the ratio between copper and barrier area constant. In aluminum lines, which are not fully encapsulated, only a bot tom titanium nitride is used for adhesion improvement. For an electrical assessment, high frequency measurements were applied to the micros trip lines based on both copper and alu minum metallization technology with silicon dioxide used as dielectric.

MICROSTRIP AND GROUNDED COPLANAR LINE
We haw measured and characterized three microstrip lines and a grounded-coplanar line. The grounded-coplanar line and two microstrip lines are realized in a copper metallization while one microstrip line is realized in a aluminum metalliza tion.

MICROSTRIP LINE IN ALUMINUM
A detailed cross-section of the 3 layer aluminum metalliza tion is shown in Fig. 5. The conductor material is standard AlSiCu and has a conductivity of a = 33 SIpm. The metal layers are embedded in silicon dioxide 8i0 2 with a relative permittivity of Sf' = 3.9. The passivation is formed by a air proof protection coat and consists of silicon nitride Si :3JY.j and has an 2:/ = / .. 5

MICROSTRIP LINE IN COPPER
The copper microstrip lines are realized in a 0.12 pm CMOS technology with six-layer copper metallization and silicon-oxide dielectric (2:1' = 3.9). Copper has a conductiv ity of a = 54 SIpm. The two topmost-layers are thick metals. The microstrip line again uses Metal 1 as ground plane. The signal lines are realized in Metal 6, which is a thick metal layer. Nevertheless. the thickness of Metal 6 of the copper process is about three times lower than Metal 3 in the alu minum process. As mentioned in Section 3 modem technolo gies have less metal thickness due to scaling. The microstrip lines in copper are realized in a similar way to the aluminum 4 microstrip line (Fig. 5). The width to height ratio of the cop per rnicrostrip lines are WIH = 1.44 and n -;H = 1.80. -'oxide@j--~~Substrat~

GROUNDED-COPLANAR LINE IN COP PER
The grounded-coplanarline is realized in the same 0.12 pm CMOS technology with six-layer copper metallization. With out a ground plane the field of the coplanar line penetrates the substrate. This causes substrate loss and is highly undesir able. To overcome the substrate loss, a ground plane shields the field against the substrate. The ground planes left and right of the line and the line itself are realized in MetalS. The ground plane under the line is realized in Metal 2. Fig. 8 shows a schematic cross-section of the grounded-coplanar line, The width to spacing ratio of the grounded-coplanar line is 11'18 = /. The width to height ratio of the ground plane to line is WI H = 3.35.

DEEM BEDDING
To extract the electrical characteristic of the waveguides from the measurement data, deembedding test structures are necessary. A calibration method with "short'iand "open'ttest structures applies correct characterization only at low fre quencies. To get accurate results at high frequencies up to 80 GHz, a thru-reflect-line (TRL) calibration is required. Fig.9(a) shows a chip micrograph of the aluminum mi crostrip line test structure with 40 um high frequency pads on left and right side to interface with ground-signal-ground probes. The total length of the microstrip line test structure is 2800 urn, Fig. 9(b) shows the chip micrograph of the cali bration test structure for deembedding.
If two-ports are connected in cascade the system can be practically defined by the transmission matrix (Tvmatrix). The measurement data are in the form of S-parameters and therefore we need the equations to get the T-matrix of a two port.
.vhere E is the unity matrix. After extracting T Line we con vert to the familiar S-parameter of the deembedded line.

PARAMETER EXTRACTION
After deembedding, we have extracted the S-parameters of the transmission lines, which describe the full electrical behavior. A figure of merit is the characteristic impedance Z and the propagation constant -, = Q + j3. where Q is the attenuation constant and 3 is the phase constant. The S-parameter measured from a lossy unmatched transmission line with characteristic impedance Z and propagation con stant i in a Zo impedance system are [6] s=2-((Z2-Z 6) sinh il 2ZZo where I is the length of the deembedded line (Fig. 9). The characteristic impedance Z in terms of the S-parameter can be written as The solutions of (9) and (10) must be chosen to be physically real.
A fundamental characteristic parameter of a waveguide is the attenuation constant Q. It represents dielectric and ohmic losses of the waveguide. (} can be calculated from (Ill where -; was derived from (9). Q = Reh}

EXPERIMENTAL RESULTS
The measurements are done on wafer-level with ground signal-ground probes. The aluminum line is measured up to 80 GHz and the copper lines are measured up to 40 GHz. Ad ditionally, the measurements of the aluminum line is com pared to simulation results.

MEASUREMENT VERSUS SIMULATION
The aluminum microstrip line was designed to match 50 D. Fig. 10 shows measured and simulated characteristic impedance Z versus frequency. The unsteady peak at 50 GHz is due to the different measurement setup in the range from 50 MHz to 50 GHz and from 50 GHz to 80 GHz.
The characteristic impedance Z is separated into the abso lute value (Fig. 10) and the phase (Fig. 11)  tion can be explained by the skin effect and the polarization losses of the dielectric. Fig. 13 shows the measured and simulated characteristic series resistance slightly increasing with frequency. The cal culated series resistance of 3.8 D/mm matches the measure ment at low frequencies. The skin-depth at 20 GHz is 0.6 pm which is about a half of the conductor height T. The series re sistance is very sensitive to measurement errors of the phase of Z.
The characteristic conductance is very sensitive to mea surement errors of the phase of Z. The values of the char acteristic conductance over frequency is in the same range as . the measurement error and therefore has not been illustrated. Fig. 14 illustrates the measured and simulated characteris tic inductance versus frequency of the microstrip line. The in ductance is slightly decreasing from DC up to 10 GHz due to the current crowding in the conductor. At frequencies higher frequency. At frequencies lower than 15 GHz the magnetic and electric-field are not in phase (-40° at 50 MHz) and the microstrip line carries a slow wave mode. At high frequen cies, the microstrip line exhibits a quasi-TEM mode. At frequencies where the microstrip line length is a multi ple of the half wave length, the S-parameter measurement is very sensitive [12]. In our case this effect causes measure ment errors at 27 GHz and 54 GHz. At these frequencies, the extracted characteristic impedance Z, the characteristic series resistance R and the characteristic capacitance C are strongly influenced due to the measurement errors, Fig. 12 shows the measured attenuation compared to sim ulations with Maxwell Field Simulator [13] and Momentum Field Simulator [14]. The small deviation between measure ment and simulation validates the parameter extraction from the S-parameter measurement. As expected. the attenuation constant 0 increases with frequency. The increasing attenua than 10 GHz the inductance is relatively constant. The characteristic capacitance plot in Fig. 15 shows a high capacitance at low frequencies and a relatively constant ca pacitance at frequencies higher than lOGHz. The capaci tance from DC to 10 GHz is reduced because the propagation mode changes from a slow wave mode to a quasi-TEM mode.

ALUMINUM LINE VERSUS COPPER LINE
More and more aluminum has been replaced by copper metallization. Copper has a much higher conductivity com pared to aluminum. Nevertheless. modem technologies have less metal thickness due to scaling (Section 3). We have real ized microstrip lines in aluminum and copper to compare the electrical behavior. On the other hand the impedance of coplanar lines is very sensitive to the spacing S (Fig. 8). This implies the need of a stable and accurate metallization process. One of the most important parameters of a waveguide is the attenuation Q. Fig. 17 shows the measured attenuation Q of the waveguides. The two copper microstrip lines have nearly the same attenuation of about 1.7 dB/mm. As not expected, the aluminum micros trip line has less attenuation than the copper lines. However, it must be reminded that the thickness of Metal 6 of the copper process is about three times lower than Metal 3 in the aluminum process. The scaling laws of equation (2) effect the attenuation of the copper lines. Low attenuation implies the need of hierarchical metallization in future process technologies (Section Z).
The skin effect is more relevant on copper lines than on aluminum lines. The skin-depth at 30 GHz in copper is 0.4 fim and in aluminum 0.54 pm. This is already the size of about a half of the conductor height T. The full advantage of copper with its high conductivity is not valid any longer at very high frequencies. Fig. 18 shows the measured characteristic series resistance of the lines. Again. the copper lines have a higher resistance than the aluminum line due to the smaller metal thickness. The skin-effect already starts at 30 GHz at the copper lines.
The characteristic conductance is very sensitive to mea surement errors of the phase of Z. The values of the char acteristic conductance over frequency is in the same range as  the measurement error and therefore has not been illustrated. Fig. 19 illustrates the measured characteristic inductance versus frequency of the waveguides. The inductance is nearly constant over the whole measured frequency range for all waveguides.
The characteristic capacitance plot in Fig.20 shows a nearly constant capacitance for all realized waveguides. The dielectric of the copper metallization and the aluminum met allization is the same (silicon-oxide, e r = 3.9). The greater distance of the copper microstrip lines to the ground plane produces a lower capacitance than on the aluminum mi crostrip line. The capacitance of the grounded-coplanar line is mainly determined by the spacing S.

CONCLUSION
In recent years, aluminum was replaced by copper due to its higher current carrying capability resulting in less power required on chips and due to its better electromigration en durance. On the other hand, continuous shrinking, which is determined by lithography limits, leads to thinner metal lay ers which have more resistance. We have shown that a copper microstrip line does not have necessarily less attenuation than an aluminum line. In next and over-next chip generations it is highly desirable to have a hierachical metallization avail able. Thick top metals are necessary to get less attenuation for global interconnects. Thin bottom metals are necessary to connect the millions of transistors. An accurate characteri zation of waveguides on silicon up to 80 GHz was presented. With the deembeding algorithm, the electrical behavior of the waveguides can be extracted. The characteristic impedance. propagation constant and telegraphers equation transmission parameter of the waveguides give a fundamental insight of its performance.

Revista da Sociedade Brasileira de Telecomunicacoes
Volume 18, Numero 1, Junho de 2003 his group for manufacturing the teststructures.