A 2.5 Gbps, 10-Lane, Low-Power, LVDS Transceiver in 28 nm CMOS Technology

Xu Bai 1,2,* , Jianzhong Zhao 1, Shi Zuo 1,2 and Yumei Zhou 1,2 1 Smart Sensing R&D Centre, Institute of Microelectronics of Chinses Academy of Science, Beijing 100029, China; zhaojianzhong@ime.ac.cn (J.Z.); zuoshi@ime.ac.cn (S.Z.); ymzhou@ime.ac.cn (Y.Z.) 2 Institute of Microelectronics, University of Chinese Academy of Sciences, 19A Yuquan Rd., Shijingshan District, Beijing 100049, China * Correspondence: baixu@ime.ac.cn; Tel.: +86-152-2230-6300


Introduction
While scaled CMOS technology continues to enhance on-chip operating speeds, the power dissipation also increases at the same time. This means that reducing power consumption is critical for battery-powered systems to extend battery life. Low voltage differential signaling (LVDS), as one of the data transmission standards, is now pervasive in communication networks and is used extensively in applications such as laptop computers [1], office imaging [2,3], and medical [4] and automotive [5,6] applications. It features a low-voltage swing (250-400 mV) and achieves a high data rate (up to several gigahertz per single pair) with less power dissipation. A typical LVDS serial link [7,8] point-to-point communication is shown in Figure 1, and involves a single transmitter (TX) and receiver (RX) pair. A current source (Is) is derived from the TX, and the output amplitude is formed by the current source flowing through the terminated resistor (R T ) to establish voltage in the input of RX. By changing the current direction, the same amplitude with the opposite polarity is created to generate the logic of zeros and ones. The simple termination, low-power, and low-noise characteristics have gradually made LVDS the technology of choice for gigabit-per-second serial transmission. In addition, the wide common mode input of LVDS makes its devices easily interoperable with other differential signaling technologies [9][10][11]. In general, the architecture of LVDS drivers is divided into fully-differential NMOS-only style [12], fully-differential PMOS-only style [13] and complementary MOS style [14][15][16]. As shown in Figure 2, all configurations consist of four MOS switches arranged in an H-bridge structure. The NMOS-only style LVDS driver, shown in Figure 2a, works well if the supply voltage (VDD) is 2.5 V or greater [17]. However, when the supply voltage is scaled down (1.8 V for 28 nm CMOS technology), it is not applicable, as there is not enough voltage headroom. According to the LVDS standard specifications [18], a 1.125-1.325 V common mode voltage range and 250-400 mV output swing of the output signals is required, which would cause the transistors (M1a and M2a) to cut off. To overcome the supply voltage headroom issues, PMOS-only (shown in Figure 2b) and complementary MOS (shown in Figure 2c) LVDS drivers need to be addressed. A benefit of PMOSonly style drivers is that they can work without the body effect. However, the inherent speed limitation in PMOS devices precludes their use in high speed data communication. To achieve the same speed as CMOS style drivers, the size of the transistors must be increased. Consequently, the area cost and power consumption will also increase. Comparing the above-mentioned LVDS drivers, the complementary MOS style driver is the optimum choice for LVDS transmission systems operating under low supply voltage, as it is not only compatible with the LVDS standard, but also faster than the other options. In this paper, a 2.5 Gbps 10-lane low-power LVDS transceiver is presented. The transceiver can operate at a data rate up to 2.5 Gbps, and is fully compatible with ANSI/TIA/EIA-644-A standards. The paper is organized as follows: Section 2 describes the architecture of the proposed LVDS transceiver, and presents some related simulation results. In Section 3, the measurement results are discussed. Finally, a summary and the conclusions are outlined in Section 4.

Architecture Design
The proposed 10-lane, low-power, LVDS transceiver is shown in Figure 3. Each lane is comprised of a receiver followed by a transmitter. It employs differential data transmission and the receiver is configured as a switched-polarity signal generator. The receiver is composed of a pre-stage common mode voltage (Vcm) shifter and a rail-to-rail comparator (COMP), while the transmitter includes a CMOS H-bridge output driver with a common mode feedback (CMFB) circuit, a high- In general, the architecture of LVDS drivers is divided into fully-differential NMOS-only style [12], fully-differential PMOS-only style [13] and complementary MOS style [14][15][16]. As shown in Figure 2, all configurations consist of four MOS switches arranged in an H-bridge structure. The NMOS-only style LVDS driver, shown in Figure 2a, works well if the supply voltage (VDD) is 2.5 V or greater [17]. However, when the supply voltage is scaled down (1.8 V for 28 nm CMOS technology), it is not applicable, as there is not enough voltage headroom. According to the LVDS standard specifications [18], a 1.125-1.325 V common mode voltage range and 250-400 mV output swing of the output signals is required, which would cause the transistors (M1a and M2a) to cut off. To overcome the supply voltage headroom issues, PMOS-only (shown in Figure 2b) and complementary MOS (shown in Figure 2c) LVDS drivers need to be addressed. A benefit of PMOS-only style drivers is that they can work without the body effect. However, the inherent speed limitation in PMOS devices precludes their use in high speed data communication. To achieve the same speed as CMOS style drivers, the size of the transistors must be increased. Consequently, the area cost and power consumption will also increase. Comparing the above-mentioned LVDS drivers, the complementary MOS style driver is the optimum choice for LVDS transmission systems operating under low supply voltage, as it is not only compatible with the LVDS standard, but also faster than the other options. In general, the architecture of LVDS drivers is divided into fully-differential NMOS-only style [12], fully-differential PMOS-only style [13] and complementary MOS style [14][15][16]. As shown in Figure 2, all configurations consist of four MOS switches arranged in an H-bridge structure. The NMOS-only style LVDS driver, shown in Figure 2a, works well if the supply voltage (VDD) is 2.5 V or greater [17]. However, when the supply voltage is scaled down (1.8 V for 28 nm CMOS technology), it is not applicable, as there is not enough voltage headroom. According to the LVDS standard specifications [18], a 1.125-1.325 V common mode voltage range and 250-400 mV output swing of the output signals is required, which would cause the transistors (M1a and M2a) to cut off. To overcome the supply voltage headroom issues, PMOS-only (shown in Figure 2b) and complementary MOS (shown in Figure 2c) LVDS drivers need to be addressed. A benefit of PMOSonly style drivers is that they can work without the body effect. However, the inherent speed limitation in PMOS devices precludes their use in high speed data communication. To achieve the same speed as CMOS style drivers, the size of the transistors must be increased. Consequently, the area cost and power consumption will also increase. Comparing the above-mentioned LVDS drivers, the complementary MOS style driver is the optimum choice for LVDS transmission systems operating under low supply voltage, as it is not only compatible with the LVDS standard, but also faster than the other options. In this paper, a 2.5 Gbps 10-lane low-power LVDS transceiver is presented. The transceiver can operate at a data rate up to 2.5 Gbps, and is fully compatible with ANSI/TIA/EIA-644-A standards. The paper is organized as follows: Section 2 describes the architecture of the proposed LVDS transceiver, and presents some related simulation results. In Section 3, the measurement results are discussed. Finally, a summary and the conclusions are outlined in Section 4.

Architecture Design
The proposed 10-lane, low-power, LVDS transceiver is shown in Figure 3. Each lane is comprised of a receiver followed by a transmitter. It employs differential data transmission and the receiver is configured as a switched-polarity signal generator. The receiver is composed of a pre-stage common mode voltage (Vcm) shifter and a rail-to-rail comparator (COMP), while the transmitter includes a CMOS H-bridge output driver with a common mode feedback (CMFB) circuit, a high- In this paper, a 2.5 Gbps 10-lane low-power LVDS transceiver is presented. The transceiver can operate at a data rate up to 2.5 Gbps, and is fully compatible with ANSI/TIA/EIA-644-A standards. The paper is organized as follows: Section 2 describes the architecture of the proposed LVDS transceiver, and presents some related simulation results. In Section 3, the measurement results are discussed. Finally, a summary and the conclusions are outlined in Section 4.

Architecture Design
The proposed 10-lane, low-power, LVDS transceiver is shown in Figure 3. Each lane is comprised of a receiver followed by a transmitter. It employs differential data transmission and the receiver is configured as a switched-polarity signal generator. The receiver is composed of a pre-stage common mode voltage (Vcm) shifter and a rail-to-rail comparator (COMP), while the transmitter includes a CMOS H-bridge output driver with a common mode feedback (CMFB) circuit, a high-speed level shifter (LS) and pre-emphasis (PE) driver. In addition, two bandgap references (BGR) are embedded in the scheme to provide proper DC bias for receivers and transmitters, respectively. In the design, the differential data are firstly addressed by the receiver, then the transmitter deals with the data and sends them out in accordance with specified requirements. Therefore, only if both the receiver and transmitter are operated properly can the transmitted signals be output. The detailed implementation of the transceiver will be expatiated in the following sections. speed level shifter (LS) and pre-emphasis (PE) driver. In addition, two bandgap references (BGR) are embedded in the scheme to provide proper DC bias for receivers and transmitters, respectively. In the design, the differential data are firstly addressed by the receiver, then the transmitter deals with the data and sends them out in accordance with specified requirements. Therefore, only if both the receiver and transmitter are operated properly can the transmitted signals be output. The detailed implementation of the transceiver will be expatiated in the following sections.

Receiver
According to LVDS specifications [18], a receiver is required to operate in a wide input common mode voltage range of 0.05-2.35 V. Therefore, with the 1.8 V supply voltage, the receiver firstly needs to achieve the common mode voltage conversion. Figure 4 shows the simplistic circuit of a pre-stage common mode voltage shifter, which includes a current regulator and an error amplifier. The error amplifier detects the common mode voltage difference between input data (INP and INN) and reference voltage (VREF) and amplifies the voltage difference to control the current regulator by injecting or extracting currents from resistors R1 and R2. As a result, voltage drops across R1 and R2 are generated, and the common mode voltage is shifted [19]. It is obvious that the shifted common mode voltage is affected by VREF. Thus, the value of VREF was set at 0.9 V for the following rail-torail comparator to obtain a higher gain. A simple rail-to-rail comparator [20,21], as shown in Figure 5, was constructed as a composite of P and NMOS pairs. The amplifier with rail-to-rail input identifies the voltage difference from the input data (OP and ON) and converts them into currents through the input trans-conductor cell (M1-M4). After this, the currents are both mirrored and summed up at the node N1, before the data is reinstituted and reshaped by the last-stage shaping buffer.

Receiver
According to LVDS specifications [18], a receiver is required to operate in a wide input common mode voltage range of 0.05-2.35 V. Therefore, with the 1.8 V supply voltage, the receiver firstly needs to achieve the common mode voltage conversion. Figure 4 shows the simplistic circuit of a pre-stage common mode voltage shifter, which includes a current regulator and an error amplifier. The error amplifier detects the common mode voltage difference between input data (INP and INN) and reference voltage (VREF) and amplifies the voltage difference to control the current regulator by injecting or extracting currents from resistors R1 and R2. As a result, voltage drops across R1 and R2 are generated, and the common mode voltage is shifted [19]. It is obvious that the shifted common mode voltage is affected by VREF. Thus, the value of VREF was set at 0.9 V for the following rail-to-rail comparator to obtain a higher gain. speed level shifter (LS) and pre-emphasis (PE) driver. In addition, two bandgap references (BGR) are embedded in the scheme to provide proper DC bias for receivers and transmitters, respectively. In the design, the differential data are firstly addressed by the receiver, then the transmitter deals with the data and sends them out in accordance with specified requirements. Therefore, only if both the receiver and transmitter are operated properly can the transmitted signals be output. The detailed implementation of the transceiver will be expatiated in the following sections.

Receiver
According to LVDS specifications [18], a receiver is required to operate in a wide input common mode voltage range of 0.05-2.35 V. Therefore, with the 1.8 V supply voltage, the receiver firstly needs to achieve the common mode voltage conversion. Figure 4 shows the simplistic circuit of a pre-stage common mode voltage shifter, which includes a current regulator and an error amplifier. The error amplifier detects the common mode voltage difference between input data (INP and INN) and reference voltage (VREF) and amplifies the voltage difference to control the current regulator by injecting or extracting currents from resistors R1 and R2. As a result, voltage drops across R1 and R2 are generated, and the common mode voltage is shifted [19]. It is obvious that the shifted common mode voltage is affected by VREF. Thus, the value of VREF was set at 0.9 V for the following rail-torail comparator to obtain a higher gain. A simple rail-to-rail comparator [20,21], as shown in Figure 5, was constructed as a composite of P and NMOS pairs. The amplifier with rail-to-rail input identifies the voltage difference from the input data (OP and ON) and converts them into currents through the input trans-conductor cell (M1-M4). After this, the currents are both mirrored and summed up at the node N1, before the data is reinstituted and reshaped by the last-stage shaping buffer. A simple rail-to-rail comparator [20,21], as shown in Figure 5, was constructed as a composite of P and NMOS pairs. The amplifier with rail-to-rail input identifies the voltage difference from the input data (OP and ON) and converts them into currents through the input trans-conductor cell (M1-M4). After this, the currents are both mirrored and summed up at the node N1, before the data is reinstituted and reshaped by the last-stage shaping buffer.

Transmitter
In this paper, the transmitter contained three parts: a high-speed level shifter, a pre-emphasis driver and an output driver. The high-speed level shifter [22,23] was introduced to achieve the different voltage domain conversion in the pre-stage of the transmitter, whose circuit is presented in Figure 6. A pair of NMOS devices (M3 and M4) receive the low-voltage input signals (Dp_L and Dn_L) and convert them into high-voltage signals through the positive feedback transistors (M1 and M2). Then, the buffer chain with several inverters reshapes the output signals under the high-voltage (VDDH) supply.  Figure 7 shows the proposed transmitter output driver based on the CMOS H-bridge structure. As Figure 7 shows, the output stage of the driver uses the PMOS and NMOS configuration. A simple common mode feedback (CMFB) circuit [24,25], with transistors M5-M8, is used to stabilize the output common mode voltage (Vcm), and is less dependent on PVT. The two differential output voltages (Voutp and Voutn) are averaged to form a common mode voltage (Vcm) by two resistors (R1 and R2), which is compared with the designed reference common mode voltage (Vbg). The difference is then amplified and converted into the common mode current to adjust the common mode voltage (Vcm). In addition, an Rc and Cc pole-zero compensation network is exploited to obtain an adequate phase margin of CMFB under the conditions created by the PVT variations. Meanwhile, a cascade current mirror (M9-M12) is utilized to provide high precision current bias at a 1.8 V voltage supply.

Transmitter
In this paper, the transmitter contained three parts: a high-speed level shifter, a pre-emphasis driver and an output driver. The high-speed level shifter [22,23] was introduced to achieve the different voltage domain conversion in the pre-stage of the transmitter, whose circuit is presented in Figure 6. A pair of NMOS devices (M3 and M4) receive the low-voltage input signals (Dp_L and Dn_L) and convert them into high-voltage signals through the positive feedback transistors (M1 and M2). Then, the buffer chain with several inverters reshapes the output signals under the high-voltage (VDDH) supply.

Transmitter
In this paper, the transmitter contained three parts: a high-speed level shifter, a pre-emphasis driver and an output driver. The high-speed level shifter [22,23] was introduced to achieve the different voltage domain conversion in the pre-stage of the transmitter, whose circuit is presented in Figure 6. A pair of NMOS devices (M3 and M4) receive the low-voltage input signals (Dp_L and Dn_L) and convert them into high-voltage signals through the positive feedback transistors (M1 and M2). Then, the buffer chain with several inverters reshapes the output signals under the high-voltage (VDDH) supply.  Figure 7 shows the proposed transmitter output driver based on the CMOS H-bridge structure. As Figure 7 shows, the output stage of the driver uses the PMOS and NMOS configuration. A simple common mode feedback (CMFB) circuit [24,25], with transistors M5-M8, is used to stabilize the output common mode voltage (Vcm), and is less dependent on PVT. The two differential output voltages (Voutp and Voutn) are averaged to form a common mode voltage (Vcm) by two resistors (R1 and R2), which is compared with the designed reference common mode voltage (Vbg). The difference is then amplified and converted into the common mode current to adjust the common mode voltage (Vcm). In addition, an Rc and Cc pole-zero compensation network is exploited to obtain an adequate phase margin of CMFB under the conditions created by the PVT variations. Meanwhile, a cascade current mirror (M9-M12) is utilized to provide high precision current bias at a 1.8 V voltage supply.  Figure 7 shows the proposed transmitter output driver based on the CMOS H-bridge structure. As Figure 7 shows, the output stage of the driver uses the PMOS and NMOS configuration. A simple common mode feedback (CMFB) circuit [24,25], with transistors M5-M8, is used to stabilize the output common mode voltage (Vcm), and is less dependent on PVT. The two differential output voltages (Voutp and Voutn) are averaged to form a common mode voltage (Vcm) by two resistors (R1 and R2), which is compared with the designed reference common mode voltage (Vbg). The difference is then amplified and converted into the common mode current to adjust the common mode voltage (Vcm). In addition, an Rc and Cc pole-zero compensation network is exploited to obtain an adequate phase margin of CMFB under the conditions created by the PVT variations. Meanwhile, a cascade current mirror (M9-M12) is utilized to provide high precision current bias at a 1.8 V voltage supply.
In addition, a pre-emphasis driver with a simple pulse-width modulation (PWM) technique [26,27] is used in the transmitter to enhance signal integrity. A simplistic circuit of this pre-emphasis driver is presented in Figure 8. The pre-emphasis driver exploits the timing relationship between signals and delay signals to establish the signal-related pulse (UP and DN), which is only enabled at the rise and fall of the signal [28,29]. During the signal transition, the pre-emphasis driver adds a current to the output node, and also extracts the current from the output node by the UP and DN pulses, so that the  Figure 9 shows the eye diagram of the transmitter after the channel, which operates at 2.5 Gbps. Figure 9a presents the simulated results of the eye diagram without a pre-emphasis driver, while the simulated results of the eye diagram with a pre-emphasis driver are shown in Figure 9b. As shown, the pre-emphasis driver is not only able to shorten the rise time but also improves the amplitude of the output signal. In addition, a pre-emphasis driver with a simple pulse-width modulation (PWM) technique [26,27] is used in the transmitter to enhance signal integrity. A simplistic circuit of this pre-emphasis driver is presented in Figure 8. The pre-emphasis driver exploits the timing relationship between signals and delay signals to establish the signal-related pulse (UP and DN), which is only enabled at the rise and fall of the signal [28,29]. During the signal transition, the pre-emphasis driver adds a current to the output node, and also extracts the current from the output node by the UP and DN pulses, so that the rise and fall time is decreased. Figure 9 shows the eye diagram of the transmitter after the channel, which operates at 2.5 Gbps. Figure 9a presents the simulated results of the eye diagram without a pre-emphasis driver, while the simulated results of the eye diagram with a preemphasis driver are shown in Figure 9b. As shown, the pre-emphasis driver is not only able to shorten the rise time but also improves the amplitude of the output signal.  In addition, a pre-emphasis driver with a simple pulse-width modulation (PWM) technique [26,27] is used in the transmitter to enhance signal integrity. A simplistic circuit of this pre-emphasis driver is presented in Figure 8. The pre-emphasis driver exploits the timing relationship between signals and delay signals to establish the signal-related pulse (UP and DN), which is only enabled at the rise and fall of the signal [28,29]. During the signal transition, the pre-emphasis driver adds a current to the output node, and also extracts the current from the output node by the UP and DN pulses, so that the rise and fall time is decreased. Figure 9 shows the eye diagram of the transmitter after the channel, which operates at 2.5 Gbps. Figure 9a presents the simulated results of the eye diagram without a pre-emphasis driver, while the simulated results of the eye diagram with a preemphasis driver are shown in Figure 9b. As shown, the pre-emphasis driver is not only able to shorten the rise time but also improves the amplitude of the output signal.  In addition, a pre-emphasis driver with a simple pulse-width modulation (PWM) technique [26,27] is used in the transmitter to enhance signal integrity. A simplistic circuit of this pre-emphasis driver is presented in Figure 8. The pre-emphasis driver exploits the timing relationship between signals and delay signals to establish the signal-related pulse (UP and DN), which is only enabled at the rise and fall of the signal [28,29]. During the signal transition, the pre-emphasis driver adds a current to the output node, and also extracts the current from the output node by the UP and DN pulses, so that the rise and fall time is decreased. Figure 9 shows the eye diagram of the transmitter after the channel, which operates at 2.5 Gbps. Figure 9a presents the simulated results of the eye diagram without a pre-emphasis driver, while the simulated results of the eye diagram with a preemphasis driver are shown in Figure 9b. As shown, the pre-emphasis driver is not only able to shorten the rise time but also improves the amplitude of the output signal.   Figure 10 shows a chip microphotograph of the 10-lane LVDS transceiver. The entire chip was fabricated with SMIC 28 nm CMOS technology and the total area was 1.46 mm 2 . The area of each TX/RX lane was 0.0333 mm 2 , where TX and RX occupy 0.0306 mm 2 and 0.0027 mm 2 , respectively. In multi-lane high-speed serial links, crosstalk and interference of lanes are important issues that deteriorate the performance of output signals. In this paper, two lanes of the transceiver shared supply voltage to improve the power integrity, and the BGR utilized a pair of individual supply voltages to provide the dependable DC bias for TX and RX, respectively. Plentiful on-chip decoupling capacitors were also inserted in the empty area to enhance signal integrity. These methods simply and effectively suppressed output jitter. multi-lane high-speed serial links, crosstalk and interference of lanes are important issues that deteriorate the performance of output signals. In this paper, two lanes of the transceiver shared supply voltage to improve the power integrity, and the BGR utilized a pair of individual supply voltages to provide the dependable DC bias for TX and RX, respectively. Plentiful on-chip decoupling capacitors were also inserted in the empty area to enhance signal integrity. These methods simply and effectively suppressed output jitter. An Agilent pulse generator 81134A was used to produce 2 31 -1 pseudorandom bit sequence (PRBS) data patterns to the receiver, while a Tektronix MSO71604C mixed signal oscilloscope was used to detect the differential output eye diagram of the transmitter. A 22-inch coupled micro-strip line on the testing PCB acted as the transmission channel, the channel loss of which is shown in Figure  11. The channel loss was 2.2 dB at 625 MHz, and 1.8 dB at 1.25 GHz. According to the measured results, the maximum data rate of the transceiver reached 2.5Gbps. Figure 12a,b shows the single lane of transmitter differential output eye diagrams with 2 31 -1 PRBS patterns and data rates of 1.25 Gbps and 2.5 Gbps. Both output swings of the two operating data rates were around 350 mV, and the root-mean-square (RMS) jitters were 5.48 ps and 3.65 ps, respectively. Figure 12c,d show transmitter differential output eye diagrams of 1.25 Gbps and 2.5 Gbps for multi- An Agilent pulse generator 81134A was used to produce 2 31 -1 pseudorandom bit sequence (PRBS) data patterns to the receiver, while a Tektronix MSO71604C mixed signal oscilloscope was used to detect the differential output eye diagram of the transmitter. A 22-inch coupled micro-strip line on the testing PCB acted as the transmission channel, the channel loss of which is shown in Figure 11. The channel loss was 2.2 dB at 625 MHz, and 1.8 dB at 1.25 GHz.

Measured Result Analysis and Discussion
multi-lane high-speed serial links, crosstalk and interference of lanes are important issues that deteriorate the performance of output signals. In this paper, two lanes of the transceiver shared supply voltage to improve the power integrity, and the BGR utilized a pair of individual supply voltages to provide the dependable DC bias for TX and RX, respectively. Plentiful on-chip decoupling capacitors were also inserted in the empty area to enhance signal integrity. These methods simply and effectively suppressed output jitter. An Agilent pulse generator 81134A was used to produce 2 31 -1 pseudorandom bit sequence (PRBS) data patterns to the receiver, while a Tektronix MSO71604C mixed signal oscilloscope was used to detect the differential output eye diagram of the transmitter. A 22-inch coupled micro-strip line on the testing PCB acted as the transmission channel, the channel loss of which is shown in Figure  11. The channel loss was 2.2 dB at 625 MHz, and 1.8 dB at 1.25 GHz. According to the measured results, the maximum data rate of the transceiver reached 2.5Gbps. Figure 12a,b shows the single lane of transmitter differential output eye diagrams with 2 31 -1 PRBS patterns and data rates of 1.25 Gbps and 2.5 Gbps. Both output swings of the two operating data rates were around 350 mV, and the root-mean-square (RMS) jitters were 5.48 ps and 3.65 ps, respectively. Figure 12c,d show transmitter differential output eye diagrams of 1.25 Gbps and 2.5 Gbps for multi- According to the measured results, the maximum data rate of the transceiver reached 2.5Gbps. Figure 12a,b shows the single lane of transmitter differential output eye diagrams with 2 31 -1 PRBS patterns and data rates of 1.25 Gbps and 2.5 Gbps. Both output swings of the two operating data rates were around 350 mV, and the root-mean-square (RMS) jitters were 5.48 ps and 3.65 ps, respectively. Figure 12c,d show transmitter differential output eye diagrams of 1.25 Gbps and 2.5 Gbps for multi-lane transmission communication. Similarly, their output swings were around 350 mV, but their performance was degraded. This is due to the lane-to-lane interference of signals and power lines, which introduced higher deterministic jitter (DJ) that deteriorated the signal integrity of the output signals. The total power dissipation of the two operating data rates were 8.72 mW and 16.51 mW at a 1.8 V power supply for each lane.
lane transmission communication. Similarly, their output swings were around 350 mV, but their performance was degraded. This is due to the lane-to-lane interference of signals and power lines, which introduced higher deterministic jitter (DJ) that deteriorated the signal integrity of the output signals. The total power dissipation of the two operating data rates were 8.72 mW and 16.51 mW at a 1.8 V power supply for each lane.  Table 1 summarizes the comparison of the performance of the previously reported LVDS transmitters. This LVDS transmitter, based on a complementary MOS H-bridge, had excellent noise immunity performance, with an RMS jitter of 3.65 ps with a data rate up to 2.5 Gbps. The proposed LVDS transmitter also had superior power consumption performance of 16.51 mW at a data rate of 2.5 Gbps, with a figure of merit (FOM) of 6.6 mW/Gbps.

Conclusions
In this paper, a 2.5 Gbps, 10-lane, low-power, LVDS transceiver was presented. In the receiver, a pre-stage common mode voltage shifter was introduced to implement the common mode voltage conversion, and a rail-to-rail comparator embedded with a shaping buffer was utilized to recover the input signal. Compared with the characteristics of previous LVDS driver architectures, a complementary MOS LVDS driver using a CMFB circuit was exploited to provide the required  Table 1 summarizes the comparison of the performance of the previously reported LVDS transmitters. This LVDS transmitter, based on a complementary MOS H-bridge, had excellent noise immunity performance, with an RMS jitter of 3.65 ps with a data rate up to 2.5 Gbps. The proposed LVDS transmitter also had superior power consumption performance of 16.51 mW at a data rate of 2.5 Gbps, with a figure of merit (FOM) of 6.6 mW/Gbps.

Conclusions
In this paper, a 2.5 Gbps, 10-lane, low-power, LVDS transceiver was presented. In the receiver, a pre-stage common mode voltage shifter was introduced to implement the common mode voltage conversion, and a rail-to-rail comparator embedded with a shaping buffer was utilized to recover the input signal. Compared with the characteristics of previous LVDS driver architectures, a complementary MOS LVDS driver using a CMFB circuit was exploited to provide the required output common mode voltage and differential output swing at 1.8 V supply voltage. In addition, a high-speed level shifter was designed for voltage domain conversion, and a pre-emphasis driver with PWM technique was employed to reduce the signal transition time. Further, the proposed LVDS transceiver was compatible with ANSI/TIA/EIA-644-A standards. The tranceiver is easy to interoperate with other differential signaling technologies, and can be embedded in other chips as an IP core, which makes it suitable for use in portable electronics. The whole circuit was fabricated with SMIC 28 nm CMOS technology, with a total chip area of 1.46 mm 2 . The measured results show that the proposed low-power LVDS was able to be properly operated at 2.5 Gbps, with an RMS jitter of 3.65 ps and an FOM of 6.6 mW/Gbps. Author Contributions: Conceptualization, X.B.; formal analysis, X.B. and J.Z.; data curation, X.B. and Y.Z.; writing-original draft preparation, X.B. and S.Z.; writing-review and editing, X.B. and S.Z.; visualization, X.B. and S.Z.; supervision, J.Z.; project administration, Y.Z.
Funding: This research was funded by National Science and Technology Major Project of China, grant number 2014ZX0302002.