RF IC semiconductor device technology

In the RF field, the requirements for performance and process are much more complex than those for digital integrated circuits themselves. Among them, power consumption, speed, and yield are the most important parameters. At the same time, RF ICs also need to take into account noise (broadband and narrowband), linearity, gain, and power efficiency. In this way, the optimized devices used in RF ICs have been continuously improved and developed. Different RF functional parts will be implemented on different semiconductor device processes. At present, the semiconductor processes used in RFICs are mainly Si, SiGe, GaAs, and InP. Their main application range in wireless communications is shown in Figure 1.

●Silicon devices: Silicon integrated circuits include silicon bipolar transistors (Si-Bipolar Transistor), silicon-complementary metal oxide semiconductor (Si-CMOS), silicon bipolar complementary metal oxide semiconductor (Bi-CMOS) or silicon germanium heterojunction bipolar transistor (SiGe HBT).

At present, the communication frequency is generally below 2 GHz. Except for power amplifiers, silicon integrated circuits have an advantage in RF/IF modules. Because of its large production capacity, silicon technology can form a single chip mixed mode IC composed of RF/IF/baseband, and can operate with a single power supply. It is far superior to GaAs devices in terms of price and integration. GaAs and silicon integrated circuits have different design methods due to different material properties. Since silicon materials do not have a semi-insulation substrate, it is equivalent to designing circuits on a high-loss substrate. In addition, the gain of the device itself is low. To achieve high-frequency electrical properties comparable to GaAs, silicon RFIC relies entirely on transistor miniaturization (such as sub-micron RF CMOS) or improvement of material structure (such as SiGe heterojunction transistors) to increase the characteristic frequency fT of the device. It is also necessary to use processes such as trench isolation to improve the isolation and Q value between circuits. The process is complicated, the number of masks is large, and the defect rate and cost are greatly increased. The high-frequency model is also difficult to master due to obvious stray effects. Currently, silicon technology is capable of RFICs above 5 GHz, but it is still insufficient for RF front-ends such as low-noise amplifiers, high-power amplifiers and switches. Therefore, devices using silicon technology will be positioned in medium-frequency modules or low-tier RF modules.

It should be pointed out that in wireless transceivers, the digital signal processing part uses the standard Si-CMOS process, which usually accounts for more than 75% of the chip area. The requirements of integration and power consumption make it impossible to implement it with other processes other than CMOS. Therefore, only by realizing the CMOS integrated RF front end can the monolithic integrated transceiver and finally the monolithic integrated mobile communication products be realized. At present, with the development of CMOS technology, its unit gain cutoff frequency has been close to the GaAs level, and some unit circuits and transceivers of RF front end realized by CMOS technology have appeared. This also makes it possible to realize the single-chip integration of mobile communication products by using CMOS technology. In addition, compared with other processes, CMOS technology has higher integration, lower cost and lower power consumption, making it the mainstream direction of RFIC development.

●GaAs devices: GaAs devices have electrical characteristics far exceeding those of silicon devices in terms of high frequency, high power, high efficiency, and low noise index. Depletion-type GaAs field-effect transistors (MESFETs) or high electron mobility transistors (HEMTs/PHEMTs) can have 80% power added efficiency (PAE) at 3 V voltage operation, making them very suitable for high-tier wireless communications that require long distances and long communication times. However, both require negative power supplies, which will increase the cost of using the product. The complicated crystal growth and gate width control of HEMT devices also affect the consistency and ease of production of the process. Enhancement mode E-mode MESFETs/HEMTs do not require negative power supplies and can maintain the excellent characteristics of their power amplifiers, but their output power will be limited. Heterogeneous bipolar transistor (HBT) is another GaAs device that does not require a negative power supply. Its power density, current drive capability and linearity all exceed those of FETs, making it suitable for designing high-power, high-efficiency, and high-linearity microwave amplifiers. HBT is the best device choice. HBT devices have advantages in phase noise, high gm, high power density, breakdown voltage and linearity. In addition, it can operate with a single power supply, thus simplifying the difficulty of circuit design and subsystem implementation. It is very suitable for the development of RF and IF transceiver modules, especially microwave signal sources and high-linearity amplifiers.

Circuit CAD Technology

For integrated circuit design, design methods and high-level computer-aided design tools are the key to success. For common VLSI, there are a series of tools including synthesis, simulation, layout design, verification, test generation, etc. to support the entire design process. However, for RFIC, there is currently no complete set of CAD tools, and the main front-end design tools are circuit-level simulation or emulation.

●Disadvantages of SPICE simulation

Conventional circuit simulation uses simulation technology represented by SPICE, which supports a variety of simulations, as shown in Table 1. However, due to the characteristics of RFIC, there are many difficulties in using this type of circuit simulation technology.

First, most of the design indicators of RFIC are indicators when the circuit is in a steady state, such as power gain, intermodulation and distortion. Time domain simulation using SPICE must go through a transient process to reach a steady state. For circuits with a longer transient process, a lot of calculations are required.

Secondly, RFIC usually has two or more signals with different frequencies or changing speeds. A typical case is a mixer, where the carrier frequency and signal frequency often differ by several orders of magnitude. Other cases such as the capture process of a PLL and the start-up process of an oscillator are very inefficient to simulate using SPICE, because the time required for simulation depends on the slowest component, while the time step depends on the fastest component.

In addition, there are distributed parasitic components such as interconnects and packaging in RFIC, which cannot be processed by SPICE. The accurate characteristics of these components must be given by electromagnetic field analysis, which is generally suitable for description in the frequency domain and cannot be directly used for analysis in the time domain.

Finally, noise is an important factor that determines the performance of IC systems, such as signal-to-noise ratio and bit error rate. However, SPICE can only perform noise analysis on linear amplifiers and when the noise source is a stationary random process. For nonlinear circuits in RFIC systems, such as mixers and oscillators, the statistical characteristics are no longer stationary because the noise is modulated by large signals, and the phase noise characteristics of the mixing noise are different from those of the oscillator. Therefore, the noise analysis method for linear circuits in SPICE cannot be used.

●RF circuit simulation technology

Due to the above reasons, traditional circuit simulation represented by SPICE cannot meet the needs of RFIC analysis. For this reason, simulation and emulation technologies specifically for RF and microwave communication circuits have been developed in the past decade.

Time domain method: Time domain simulation generally solves the circuit time domain differential equations under the assumption that the steady-state response of the circuit is periodic, that is, v(0)=v(T), where v is the node voltage vector, T is the period, v(0) is the initial vector of the node voltage at time zero, and v(T) is the node voltage vector at time T, and then finds the initial state v(0) that makes the equation have a periodic solution. For circuits whose excitation signals are periodic signals, the period T is a known quantity, but for oscillating circuits, its period is generally unknown, so in addition to determining v(0), the period T must also be determined.

The most commonly used method to solve the above equations is the Newton test method. Its basic principle is: assuming that the corresponding period T of the circuit is known, in a certain initial state, perform traditional circuit transient analysis on the circuit within the period T to determine whether v(0)=v(T) is satisfied. If not, set v(0)=v(T) and perform transient analysis again. Repeat this iteration until an initial state that satisfies v(0)=v(T) is found. Figure 2 shows the v(t) waveform after 5 Newton iterations.

In the above process, a large number of matrix operations need to be performed, so this limits the scale of the circuit. The current simulation generally does not exceed 300 nodes.

The trial-and-error method is a method in the time domain. The strength of the circuit nonlinearity or whether the signal is close to a sine wave does not affect the scale of the equation and the amount of memory. The convergence of the iteration depends on the degree of nonlinearity between v(T) and v(0), rather than the nonlinearity of the circuit itself. Therefore, it can also converge for some strongly nonlinear circuits. Its disadvantage is that it is difficult to handle discrete components. In the time domain, in order to accurately calculate the distortion, it is necessary to select appropriate simulation tolerances and algorithms.

Harmonic balance method: Harmonic balance is a method for finding the steady-state response of a circuit in the frequency domain. First, the signal is expressed in the form of Fourier expansion, and the KCL equations are listed for each harmonic component at the node. The differential equations in the time domain are converted into algebraic equations in the frequency domain, and then the Fourier coefficients are solved using Newton iteration. It should be noted that since the characteristics of nonlinear elements are expressed in the time domain, their calculations must first be performed in the time domain, and then transformed into the frequency domain using Fourier transform. In order to calculate the nonlinear resistor current and nonlinear capacitor charge in the time domain, the excitation signal V (ω) must first be converted into the time domain using inverse Fourier transform.

The harmonic balance method is essentially a nonlinear analysis method in the frequency domain, which is suitable for approximate sinusoidal steady-state analysis of circuits with weak nonlinearity, such as distortion and intermodulation analysis of amplifiers. When the nonlinearity of the circuit is strong, many harmonic components of the fundamental wave are used to simulate the distorted sinusoidal signal. The greater the distortion, the more harmonics are used, which will increase the scale of the equation. Another difficulty in nonlinearity is that it is more difficult to converge during iteration.

Conclusion

The development direction of RF integrated circuits is higher frequency application range and wider bandwidth, which requires the continuous development of new semiconductor technology processes and more accurate and reliable CAD technical support in design.