A 40 Gbps low-power inverter-based CNTFET optical receiver Authors: Nazanin Zamani, Mehdi Amoon, Zahra Alaie, Mahmoud Daneshvar Farzanegan Published: 2026 ABSTRACT This paper presents a low-power 40 Gbps optical receiver implemented using 32 nm carbon nanotube field-effect transistor (CNTFET) technology. The proposed receiver architecture consists of a modified inverter-based transimpedance amplifier (TIA) followed by cascaded limiting amplifiers (LAs). Using the gm/ID design methodology, the TIA is optimized to achieve a balanced trade-off among gain, bandwidth, noise, and power consumption. Simulation results indicate that the proposed TIA achieves a transimpedance gain of 47.6 dBΩ, a − 3 dB bandwidth of 28.4 GHz, and an input-referred noise current of 15.5 pA/√Hz, while consuming only 183 µW and occupying an area of 3.93 μm². The complete receiver provides an overall gain of 77.8 dBΩ, a − 3 dB bandwidth of 29.5 GHz, and a total power consumption of 1377 µW within a layout area of 49.19 μm². In addition, analytical predictions derived from the gm/ID framework are in close agreement with the simulation results, thereby validating the adopted design methodology. Compared with conventional CMOS-based implementations, the proposed CNTFET-based receiver demonstrates notable improvements in power efficiency, noise performance, and integration density, indicating its potential for next-generation optical interconnects and communication systems.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-54513-0. Introduction The optical communication system illustrated in Fig. 1 consists of three main building blocks: a transmitter, an optical fiber link, and a receiver. In the transmitter, an externally modulated laser (EML) maps the input electrical data onto an optical carrier. The modulated optical signal then propagates through the optical fiber and reaches the receiver, where a photodiode detects the incident optical signal and converts it into a corresponding photocurrent1. This photocurrent is subsequently converted into a voltage signal by a TIA. The resulting voltage is further amplified by a LA to provide the amplitude levels required for subsequent digital signal processing2. Fig. 1Block diagram of the optical communication system2. Block diagram of the optical communication system2. TIAs are critical components in optical receivers, and different TIA architectures offer distinct trade-offs among gain, bandwidth, noise, stability, power consumption, and implementation complexity. Therefore, optimizing the TIA topology is essential for improving the overall performance of high-speed optical receiver systems. This study focuses on the design and optimization of a TIA architecture aimed at enhancing gain and bandwidth while maintaining low power consumption. A resistive-feedback TIA offers structural simplicity and low-noise operation; however, its high-frequency performance is limited by parasitic capacitances and stability constraints3. A common-gate TIA also provides a relatively simple topology, but its bandwidth and noise performance can be constrained by input impedance requirements and parasitic effects2. Fully differential TIAs can achieve high gain, wide bandwidth, improved linearity, and strong common-mode rejection, although these advantages are typically obtained at the cost of increased power consumption and circuit complexity4. The regulated-cascode (RGC) TIA provides low input impedance and wide output swing while maintaining high output resistance5. Inverter-based TIAs are attractive because of their high effective transconductance and compact implementation; however, their bandwidth may be limited by intrinsic capacitances6. To mitigate these limitations, cascaded-inverter TIA architectures have been proposed to improve gain, bandwidth, and noise performance, albeit generally at the expense of higher power dissipation7. In this work, a modified inverter-based TIA is introduced as the primary amplification stage, with particular emphasis on bandwidth enhancement and power-efficient operation. The selection of an appropriate TIA architecture depends on application-specific requirements, including transimpedance gain, bandwidth, noise performance, and power consumption. For reliable operation in high-speed optical receivers, the TIA bandwidth is typically designed to be approximately 70% of the target data rate8. Several bandwidth-enhancement techniques have therefore been developed to address the intrinsic frequency limitations of TIA circuits. Stagger tuning and T-coil peaking extend the bandwidth through inductive peaking, but they require substantial chip area because of the passive inductive components9,10. The Cherry–Hooper topology improves bandwidth by employing local feedback; however, it typically requires a higher supply voltage2,11. Capacitive degeneration introduces a compensating zero that shifts the dominant output pole to higher frequencies, although this approach can increase thermal noise12. Other techniques, such as fT doublers and negative capacitance, can enhance high-frequency response, but they may lead to increased power consumption or be limited by capacitor-sizing constraints13,14. Among these techniques, the active-inductor approach is particularly attractive because it emulates inductive behavior and resonates with the load capacitance, thereby extending the bandwidth without requiring large passive inductors15. In this work, this technique is adopted to achieve bandwidth enhancement under low-power operating conditions. In optical receiver systems, the low-amplitude output voltage generated by the TIA is further amplified by an LA to reach signal levels suitable for digital processing. The LA must provide sufficient gain to suppress the effect of noise contributed by subsequent stages while preserving the bandwidth required for high-speed operation. When multiple LA stages are cascaded, each stage is commonly designed with a bandwidth of approximately 80% of the target data rate to compensate for cumulative bandwidth degradation. In addition to gain and bandwidth, the input-referred noise of the LA is a key design parameter because a wider bandwidth generally increases the integrated noise. The total input-referred noise current of the cascaded TIA–LA receiver depends on the noise bandwidth and the equivalent input noise contributions of both stages. Consequently, minimizing the LA input-referred noise voltage is essential for reducing the overall receiver noise2. As electronic devices continue to scale into the nanometer regime, conventional metal–oxide–semiconductor field-effect transistor (MOSFET) technology faces several fundamental limitations, including short-channel effects, increased leakage current, high electric fields, lithographic constraints, and quantum-confinement phenomena16. To address these challenges, CNTFETs have emerged as promising alternatives because of the exceptional electrical and structural properties of carbon nanotubes17. CNTFETs offer several advantages over conventional MOSFETs, including improved electrostatic gate control, high carrier mobility for both n-type and p-type devices, near-ballistic transport, high transconductance, large on/off current ratios, and tunable threshold voltage characteristics18,19–25. Their excellent electrical and thermal conductivities can further improve device- and circuit-level performance25. Despite fabrication-related challenges and sensitivity to structural defects26, CNTFETs can be compatible with standard CMOS-oriented fabrication flows, enabling the adaptation of established MOSFET design methodologies and tools while reducing implementation complexity27,28. The remainder of this paper is organized as follows. Section 2 presents the overall architecture and design methodology of the proposed modified TIA, LA, and complete optical receiver. It also introduces the active-inductor technique for bandwidth enhancement and analyzes the input-referred noise characteristics of the modified TIA using an equivalent noise circuit model. Section 3 describes the gm/ID design methodology and demonstrates its application to the optimization of the gain, bandwidth, and noise performance of the proposed TIA. Analytical formulations and simulation results are then provided to validate the effectiveness of the design approach. Section 4 discusses the simulation results for the amplifiers and the complete optical receiver, including performance comparisons with previously reported 40 Gbps designs. Finally, Sect. 5 concludes the paper. Modified circuit architecture and design Conventional inverter-based TIA topology Recent studies have shown that the inverter, although traditionally regarded as a fundamental digital building block, can also operate effectively in analog applications29,30. When both transistors operate in the saturation region, small variations in the input voltage can induce substantial changes at the output31. Under this condition, the inverter behaves as a voltage amplifier and has therefore been adopted as the core element in a variety of analog circuit implementations32. Simultaneous conduction of the n-channel and p-channel transistors increases the effective transconductance, thereby extending the bandwidth and enabling high-speed amplification in optical receiver applications6. Inverter-based architectures also offer favorable noise characteristics because the input-referred noise contribution is shared by both transistors33,34,35. In addition, their inherently low input resistance makes them well suited for wideband applications36–38. Figure 2 shows a conventional inverter-based TIA employing a resistive feedback network. Despite its structural simplicity, this topology suffers from several limitations, including limited DC gain and increased power consumption at higher supply voltages. Fig. 2Conventional inverter-based TIA32. Conventional inverter-based TIA32. The low-frequency transimpedance gain of the conventional inverter-based TIA can be expressed as1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Z}_{T}=\frac{{V}_{out}}{{I}_{PD}}={R}_{in}{A}_{V}$$\end{document} Based on small-signal analysis, the voltage gain AV is given by2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{A}_{V}=\frac{{V}_{out}}{{V}_{in}}=\frac{\left(\frac{1}{{R}_{F}}-{g}_{m1}-{g}_{m2}\right)}{\left(\frac{1}{{R}_{F}}+\frac{1}{{r}_{ds1}}+\frac{1}{{r}_{ds2}}\right)}$$\end{document} By considering the Miller effect, the input resistance Rin is expressed as3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{in}=\frac{{R}_{F}}{1-{A}_{V}}$$\end{document} Because the input pole is located at low frequency, the desired bandwidth must be achieved by properly placing the output pole. Assuming that the output pole is dominant, the − 3 dB bandwidth is defined as4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{-3dB}=\frac{1}{2\pi\:{C}_{out}}\times\:\frac{1}{{R}_{out}}\approx\:\frac{1}{2\pi\:{C}_{L}{R}_{F}}$$\end{document} The output capacitance Cout is given by5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{out}={C}_{L}+{C}_{dg1}\left(1-{A}_{V}^{-1}\right)+{C}_{db1}+{C}_{dg2}\left(1-{A}_{V}^{-1}\right)+{C}_{db2}\approx\:{C}_{L}$$\end{document} Similarly, the output resistance Rout, while accounting for the Miller effect, can be written as6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{out}={r}_{ds1}\parallel{r}_{ds2}\parallel\frac{{R}_{F}}{\left(1-{A}_{V}^{-1}\right)}\approx\:{R}_{F}$$\end{document} The input-referred noise of the conventional inverter-based TIA is formulated as7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{I}_{n,in}^{2}}={G}_{in}^{2}\overline{{V}_{n,in}^{2}}=\left(\frac{1}{{R}_{in}^{2}}+{C}_{in}^{2}{\omega\:}^{2}\right)\frac{4KT\gamma\:\left[{g}_{m1}+{g}_{m2}+{g}_{m{R}_{F}}\right]\times\:{R}_{out}^{2}}{{A}_{v}^{2}}$$\end{document} In Eq. (7), AV, Rin, and Rout are defined by Eqs. (2), (3), and (6), respectively. The total input capacitance Cin, representing the capacitance seen at the input node, is given by8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{in}={C}_{PD}+{C}_{gd1}\left(1-{A}_{V}\right)+{C}_{gs1}+{C}_{gb1}+{C}_{gd2}\left(1-{A}_{V}\right)+{C}_{gs2}+{C}_{gb2}\approx\:{C}_{PD}\:$$\end{document} Modified transimpedance amplifier Figure 3 illustrates the proposed modified cascode inverter-based TIA, which consists of CNTFETs CNT1–CNT4. The circuit incorporates cascoding, resistive feedback, diode-connected loading, and active-inductor-based peaking techniques. The cascode configuration increases the inverter output impedance to approximately gm2rds2rds1∥gm3rds3rds4, thereby improving the DC gain compared with a conventional inverter-based TIA while maintaining nearly the same power consumption37. CNT5 operates as a feedback resistor with an equivalent resistance of rds5, which reduces the output resistance and low-frequency gain and consequently extends the bandwidth6,38. CNT6 is implemented as a diode-connected device and behaves as a small-signal resistance of approximately 1/gm6, further lowering the output resistance32. The active inductor formed by CNT7 and CNT8 introduces controlled inductive peaking by generating an equivalent inductance that resonates with the output capacitance, thereby extending the − 3 dB bandwidth15. Although active inductors generally exhibit higher noise, lower quality factor, and greater sensitivity to bias and process variations than passive inductors, they provide important advantages for this design, including electronic tunability, compact area, and compatibility with fully integrated multi-GHz operation. In particular, the peaking response can be adjusted through transistor sizing and bias optimization, while the use of only two CNTFET devices substantially reduces the chip area compared with an on-chip passive inductor. To minimize power consumption, the TIA is implemented as a single-stage amplifier. The use of active devices also reduces the occupied area and simplifies integration. Fig. 3Modified cascode inverter-based TIA topology. Modified cascode inverter-based TIA topology. Fig. 4Active inductor implementation: (a) small-signal model and (b) frequency-dependent output impedance15. Active inductor implementation: (a) small-signal model and (b) frequency-dependent output impedance15. Figure 4 shows the small-signal model and the frequency-dependent output impedance of the active inductor15. The output impedance at the node indicated in Fig. 4(a) is expressed as159\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Z}_{OUT}=\frac{{r}_{ds7}\:{C}_{gs8}\:S+1}{{C}_{gs8}\:S+{g}_{m8}}$$\end{document} The zero and pole introduced by the active inductor are given by1510\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\omega\:}_{z}=-\frac{1}{{r}_{ds7}\:{C}_{gs8}}$$\end{document}11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\omega\:}_{p}=-\frac{{g}_{m8}}{{C}_{gs8}\:}$$\end{document} At low frequencies, Cgs8 presents a high impedance, and CNT8 behaves as a diode-connected device; therefore, ZOUT ≈ 1/gm8. At high frequencies, Cgs8 provides a low-impedance path that AC-grounds the gate of CNT8, leading to ZOUT ≈ rds7. When 1/gm8 < rds7, equivalently ωz < ωp, the circuit exhibits inductive behavior over the frequency range ωz < ω < ωp. The equivalent inductance in this region is expressed as15. 12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{L}_{eq}=\left({r}_{ds7}-\frac{1}{{g}_{m8}}\right)\frac{{C}_{gs8}}{{g}_{m8}}\underrightarrow{\:\:\:\:for\:{r}_{ds7}\gg\:1/{g}_{m8}\:\:\:\:}{L}_{eq}=\frac{{{r}_{ds7}C}_{gs8}}{{g}_{m8}}$$\end{document} Fig. 5Small-signal equivalent circuit of the modified cascode inverter-based TIA. Small-signal equivalent circuit of the modified cascode inverter-based TIA. Figure 5 presents the small-signal equivalent circuit of the proposed modified TIA. Applying KCL at the output node, the input node, and the drain nodes of CNT1 and CNT4 yields13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}\:\frac{{V}_{out}-{V}_{in}}{{r}_{ds5}}&+{g}_{m2}{V}_{gs2}+\frac{{V}_{out}-{V}_{in}+{V}_{gs2}}{{r}_{ds2}}+{g}_{m3}{V}_{gs3}+\frac{{V}_{out}-{V}_{in}+{V}_{gs3}}{{r}_{ds3}}\\&+\frac{{V}_{out}}{\frac{1}{{g}_{m6}}}+\frac{{V}_{out}-{V}_{gs8}}{{r}_{ds7}}={g}_{m8}{V}_{gs8}\end{aligned}$$\end{document}14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{I}_{PD}+\frac{{V}_{in}-{V}_{gs2}}{{r}_{ds1}}+\frac{{V}_{in}-{V}_{gs3}}{{r}_{ds4}}=\frac{{V}_{out}-{V}_{in}}{{r}_{ds5}}$$\end{document}15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{g}_{m2}{V}_{gs2}+\frac{{V}_{out}-{V}_{in}+{V}_{gs2}}{{r}_{ds2}}=\frac{{V}_{in}-{V}_{gs2}}{{r}_{ds1}}$$\end{document}16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{g}_{m3}{V}_{gs3}+\frac{{V}_{out}-{V}_{in}+{V}_{gs3}}{{r}_{ds3}}=\frac{{V}_{in}-{V}_{gs3}}{{r}_{ds4}}$$\end{document} By expressing Vgs2 and Vgs3 in terms of Vin and Vout and substituting them into the output-node equation, the overall voltage gain is obtained as17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{A}_{V}=\frac{{V}_{out}}{{V}_{in}}=\frac{\frac{\left({g}_{m2}+\frac{1}{{r}_{ds2}}\right)\left(\frac{1}{{r}_{ds1}}+\frac{1}{{r}_{ds2}}\right)}{\left({g}_{m2}+\frac{1}{{r}_{ds2}}+\frac{1}{{r}_{ds1}}\right)}+\frac{\left({g}_{m3}+\frac{1}{{r}_{ds3}}\right)\left(\frac{1}{{r}_{ds3}}+\frac{1}{{r}_{ds4}}\right)}{\left({g}_{m3}+\frac{1}{{r}_{ds3}}+\frac{1}{{r}_{ds4}}\right)}-\left(\frac{1}{{r}_{ds2}}+\frac{1}{{r}_{ds3}}+\frac{1}{{r}_{ds5}}\right)}{\frac{\frac{1}{{r}_{ds2}}\left({g}_{m2}+\frac{1}{{r}_{ds2}}\right)}{\left({g}_{m2}+\frac{1}{{r}_{ds2}}+\frac{1}{{r}_{ds1}}\right)}+\frac{\frac{1}{{r}_{ds3}}\left({g}_{m3}+\frac{1}{{r}_{ds3}}\right)}{\left({g}_{m3}+\frac{1}{{r}_{ds3}}+\frac{1}{{r}_{ds4}}\right)}-\left(\frac{1}{{r}_{ds2}}+\frac{1}{{r}_{ds3}}+\frac{1}{{r}_{ds5}}+{g}_{m6}-{g}_{m8}\right)}$$\end{document} Assuming gm6 = gm8 and sufficiently large values of gm2 and gm3, the terms 1/rds2 and 1/rds3 can be neglected. Under these assumptions, (17) simplifies to18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{A}_{V}=\frac{{V}_{out}}{{V}_{in}}=\frac{\frac{A}{C}\left({g}_{m2}\right)+\frac{B}{D}\left({g}_{m3}\right)-E}{\frac{\left({g}_{m2}\right)}{C{r}_{ds2}}+\frac{\left({g}_{m3}\right)}{D{r}_{ds3}}-E}$$\end{document} Where A = (1/rds1 + 1/rds2), B = (1/rds3 + 1/rds4), C = (gm2 + 1/rds1), D = (gm3 + 1/rds4), and E = (1/rds2 + 1/rds3 + 1/rds5) When A/C and B/D are sufficiently large, the proposed amplifier can provide a higher voltage gain than the conventional inverter-based TIA. The frequency response contains two poles and one zero. The input pole is expressed as19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\omega\:}_{p,in}=-\frac{1}{{R}_{in}{C}_{in}}$$\end{document} where the equivalent input resistance and capacitance are given by20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{in}=\frac{{r}_{ds5}}{1-{A}_{V}}$$\end{document}21\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}\:{C}_{in}&={C}_{PD}+{C}_{gs2}+{C}_{gd2}\left(1-{A}_{V}\right)+{C}_{gb2}+{C}_{gs3}+{C}_{gd3}\left(1-{A}_{V}\right)+{C}_{gb3}\\&+{C}_{sg5}+{C}_{sb5}\approx\:{C}_{PD}\end{aligned}$$\end{document} The dominant output pole is22\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\omega\:}_{p,out}=-\frac{1}{{R}_{out}{C}_{out}}$$\end{document} where the equivalent output resistance and capacitance are calculated as23\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{out}={g}_{m2}{r}_{ds2}{r}_{ds1}\parallel{g}_{m3}{r}_{ds3}{r}_{ds4}\parallel\frac{{r}_{ds5}}{1-{A}_{V}^{-1}}\parallel\frac{1}{{g}_{m6}}\parallel\frac{1}{{g}_{m8}}\approx\:{r}_{ds5}$$\end{document}24\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}\:{C}_{out}&={C}_{L}+{C}_{dg2}\left(1-{A}_{V}^{-1}\right)+{C}_{db2}+{C}_{dg3}\left(1-{A}_{V}^{-1}\right)+{C}_{db3}+{C}_{dg5}+{C}_{db5}+{C}_{db6}\\&+{C}_{sg8}+{C}_{sb8}\approx\:{C}_{L}\end{aligned}$$\end{document} Accordingly, the − 3 dB bandwidth is given by f−3dB = ωp, out/2π. Assuming the same effective output capacitance, CL, the proposed design can achieve a wider bandwidth if the effective output resistance is reduced relative to that of the conventional inverter-based TIA. Figure 6 shows the thermal-noise equivalent circuit of the modified TIA. The input-referred noise determines the minimum detectable signal for a specified signal-to-noise ratio (SNR). It is analytically expressed as32. 25\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{I}_{n,in}^{2}}={G}_{in}^{2}\overline{{V}_{n,in}^{2}}$$\end{document} Fig. 6Thermal noise equivalent circuit of the modified cascode inverter-based TIA. Thermal noise equivalent circuit of the modified cascode inverter-based TIA. The output thermal-noise voltage is given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{V}_{n,out}^{2}}=\overline{{I}_{n,out}^{2}}\times\:{\left|{R}_{out}\right|}^{2}$$\end{document} where Rout ≈ rds5. The total output thermal-noise current and voltage are26\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{I}_{n,out}^{2}}=\overline{{I}_{n,CNT2}^{2}}+\overline{{I}_{n,CNT3}^{2}}+\overline{{I}_{n,CNT5}^{2}}+\overline{{I}_{n,CNT6}^{2}}+\overline{{I}_{n,CNT8}^{2}}$$\end{document}27\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}\:\overline{{I}_{n,out}^{2}}&=4KT\gamma\:{g}_{m2}+4KT\gamma\:{g}_{m3}+4KT\gamma\:{g}_{m5}+4KT\gamma\:{g}_{m6}+4KT\gamma\:{g}_{m8}\\&=4KT\gamma\:\left[{g}_{m2}+{g}_{m3}+{g}_{m5}+{g}_{m6}+{g}_{m8}\right]\approx\:4KT\gamma\:\left[{g}_{m6}+{g}_{m8}\right]\end{aligned}$$\end{document}28\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{V}_{n,out}^{2}}=\overline{{I}_{n,out}^{2}}\times\:{\left|{R}_{out}\right|}^{2}=4KT\gamma\:\left[{g}_{m6}+{g}_{m8}\right]\times\:{\left({R}_{out}\right)}^{2}$$\end{document} The input-referred thermal-noise voltage is obtained by dividing the output noise voltage by the squared voltage gain. Consequently, the corresponding input-referred noise current is29\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{V}_{n,in}^{2}}=\frac{\overline{{V}_{n,out}^{2}}}{{A}_{v}^{2}}=\frac{4KT\gamma\:\left[{g}_{m6}+{g}_{m8}\right]\times\:{\left({R}_{out}\right)}^{2}}{{A}_{v}^{2}}$$\end{document}30\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{I}_{n,in}^{2}}={G}_{in}^{2}\overline{{V}_{n,in}^{2}}=\left(\frac{1}{{R}_{in}^{2}}+{C}_{in}^{2}{\omega\:}^{2}\right)\frac{4KT\gamma\:\left[{g}_{m6}+{g}_{m8}\right]\times\:{\left({R}_{out}\right)}^{2}}{{A}_{v}^{2}}$$\end{document} Assuming identical input resistance and capacitance, comparison with the conventional inverter-based TIA indicates that the proposed topology can reduce the input-referred noise primarily through its higher voltage gain. Limiting amplifier stage To provide the overall voltage gain required in the optical receiver, three LA stages are cascaded. Figure 7 shows the block diagram and circuit-level implementation of this structure. Each LA stage is realized using a conventional differential-pair topology with active-inductor loads to provide sufficient bandwidth while maintaining low power consumption. Fig. 7(a) Block diagram and (b) circuit-level implementation of the three-stage cascaded limiting amplifier. (a) Block diagram and (b) circuit-level implementation of the three-stage cascaded limiting amplifier. Under small-signal conditions, the low-frequency voltage gain of a single LA stage can be approximated as31\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{A}_{V}=\frac{{V}_{out}}{{V}_{in}}={-g}_{m1}\times\:\left({r}_{ds1}\parallel\frac{1}{{g}_{m3}}\right)$$\end{document} where gm1 is the transconductance of the input transistor, rds1 is the small-signal output resistance of the input transistor, and 1/gm3 represents the equivalent load resistance at low frequencies. Since the limiting amplifier consists of three cascaded stages, the overall gain is obtained from the product of the gains of the individual stages. Moreover, the frequency response of each stage is modeled as a single-pole block, in which the output pole is assumed to be dominant. Accordingly, the − 3 dB bandwidth of each stage can be expressed as32\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{-3dB}=\frac{1}{2\pi\:{C}_{out}{R}_{out}}$$\end{document} where33\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{out}={r}_{ds1}\parallel\frac{1}{{g}_{m3}}$$\end{document} and34\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{out}={C}_{dg1}+{C}_{db1}+{C}_{dg3}+{C}_{db3}+{C}_{dg4}+{C}_{db4}+{C}_{Load}$$\end{document} In cascaded multi-stage structures, cascading several similar LA stages reduces the overall bandwidth of the signal chain. Therefore, the bandwidth of each stage must be sufficiently large. In addition, without the active-inductor technique, the bandwidth would be mainly limited by the RC time constants of the internal nodes. By contrast, the active-inductor load partially compensates for the effect of parasitic capacitances and thus increases the effective bandwidth. From a noise standpoint, the main noise sources in the limiting amplifier include transistor channel thermal noise, flicker noise, and noise referred from preceding stages. Due to the high operating frequency, flicker noise is of less importance than thermal noise in a first-order analysis. The input-referred noise current density of each stage can be generally expressed as35\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{I}_{n,in}^{2}}=\left(\frac{1}{{R}_{in}^{2}}+{C}_{in}^{2}{\omega\:}^{2}\right)\frac{{\left|{R}_{out}\right|}^{2}\times\:\overline{{I}_{n,out}^{2}}}{{A}_{v}^{2}}$$\end{document} where the output noise current is36\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{I}_{n,out}^{2}}=\overline{{I}_{n,CNT1}^{2}}+\overline{{I}_{n,CNT3}^{2}}++\overline{{I}_{n,CNT4}^{2}}=4KT\gamma\:{g}_{m1}+4KT\gamma\:{g}_{m3}+4KT\gamma\:{g}_{m4}$$\end{document} In these expressions, for the first LA stage, Rin and Cin correspond to the output resistance and output capacitance of the TIA, respectively, whereas for the second and third stages, they correspond to the input resistance and input capacitance of the preceding LA stage. Furthermore, in cascaded structures, the earlier stages contribute more significantly to the total noise, because the noise components of later stages are suppressed by the square of the gains of the preceding stages. Optical receiver system The complete architecture of the proposed optical receiver is shown in Fig. 8. In the analysis presented in this study, the photodiode is represented by a differential structure based on the circuit model, in which the photodiode current is applied differentially to the two TIA inputs. This representation is architectural in nature and does not necessarily imply the use of a commercial differential photodiode. The receiver consists of two modified cascode inverter-based TIA stages followed by three cascaded LA stages. The TIA stages are implemented in a fully differential configuration to suppress common-mode noise and reduce the overall noise contribution. Such a differential implementation is particularly advantageous for low-noise optical detection systems2. Fig. 8Block diagram of the low-noise optical receiver system employing the circuit-model-based photodiode representation, followed by a modified cascode inverter-based TIA and three cascaded LA stages2. Block diagram of the low-noise optical receiver system employing the circuit-model-based photodiode representation, followed by a modified cascode inverter-based TIA and three cascaded LA stages2. gm/ID design methodology The gm/ID methodology provides a systematic framework for optimizing analog circuit performance by using the transconductance-to-drain-current ratio as a key design parameter39. This approach simplifies the design procedure, reduces design-variable interdependence, and enables a trade-off between power efficiency and circuit performance40. Although this methodology is well established for MOSFET-based circuits, its application to CNTFET-based designs introduces additional challenges due to the quantized dimensions of carbon nanotubes41. Figure 9 illustrates the proposed gm/ID-based design flowchart for the modified TIA, including the steps required to achieve the target transimpedance gain, bandwidth, and input-referred noise. The procedure begins by selecting suitable gm/ID ratios and drain currents for each CNTFET according to the device function and the constraints listed in Table 1. These parameters are then substituted into the analytical expressions for gain, bandwidth, and input-referred noise derived within the gm/ID framework. The design is iteratively refined until the target specifications are satisfied. Once the design objectives are met, the bias voltages and the number of nanotubes in each CNTFET are determined to realize the required conductivity and drain current. The complete circuit is subsequently simulated to evaluate the transistor operating regions, inductive peaking behavior, eye-diagram quality, and total power consumption. If the simulated performance deviates from the design targets, the parameters are adjusted and the procedure is repeated. The design process is completed when the simulation results satisfy the specified criteria. The transimpedance gain, bandwidth, and input-referred noise of the modified TIA can be expressed within the gm/ID methodology as follows:37\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}\:{Z}_{T}&={R}_{in}{A}_{V}={R}_{in}\\&\times\:\frac{\left[\left({\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{2}+A\right)/\left(\frac{1}{C}{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{2}+1\right)\right]+\left[\left({\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{3}+B\right)/\left(\frac{1}{D}{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{3}+1\right)\right]-\left(A{I}_{D2}+B{I}_{D3}\right)}{\left[\left(A{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{2}+{A}^{2}\right)/\left({\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{2}+C\right)\right]+\left[\left(B{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{3}+{B}^{2}\right)/\left({\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{3}+D\right)\right]-\left(A{I}_{D2}+B{I}_{D3}\right)}\end{aligned}$$\end{document} where A = λ2, B = λ3, C = λ1 + λ2, D = λ3 + λ4, and λ = 1/rdsID. 38\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{-3dB}=\frac{1}{2\pi\:{C}_{L}}\times\:E$$\end{document} where E = 1/rds5. The low-frequency input-referred noise, for s→0, is expressed as39\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{I}_{n,in}^{2}}=\left(\frac{1}{{R}_{in}^{2}}\right)\frac{4KT\gamma\:\left({I}_{D6}{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{6}+{I}_{D8}{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{8}\right)\times\:{\left({R}_{out}\right)}^{2}}{{A}_{v}^{2}}$$\end{document} Table 1 summarizes the required parameter ranges and the simulated gm/ID values for each transistor in the modified cascode inverter-based TIA obtained using the proposed methodology. Table 1Required and simulated gm/ID values for transistors in the modified cascode inverter-based TIA. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{1}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{2}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{3}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{4}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{5}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{6}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{7}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{8}$$\end{document} Required value[10–25][10–25][10–25][10–25][1-5k][5–10][1–5][5–10]Simulated (℧/A)19.5920.6620.21202.95k9.272.669.27 Table 1Required and simulated gm/ID values for transistors in the modified cascode inverter-based TIA. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{1}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{2}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{3}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{4}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{5}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{6}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{7}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left[\frac{{g}_{m}}{{I}_{D}}\right]}_{8}$$\end{document} Required value[10–25][10–25][10–25][10–25][1-5k][5–10][1–5][5–10]Simulated (℧/A)19.5920.6620.21202.95k9.272.669.27 Required and simulated gm/ID values for transistors in the modified cascode inverter-based TIA. Table 2 compares the target specifications, simulation results, and the corresponding gm/ID-based estimates. The results indicate reasonable agreement between the analytical calculations and circuit-level simulations. The observed discrepancy in bandwidth is attributed to the low-frequency approximations used in the analytical derivation. Fig. 9Proposed gm/ID-based design methodology. Proposed gm/ID-based design methodology. Table 2Comparison of target specifications, simulation results, and gm/ID-based estimates for the modified cascode inverter-based TIA.gm/IDAV (dBΩ)f−3dB (GHz)IRN (A/√Hz)Required value> 40≈ 28< 20 pSimulated (TIA)47.628.415.5 pgm/ID methodology50.6146.97.9 p Table 2Comparison of target specifications, simulation results, and gm/ID-based estimates for the modified cascode inverter-based TIA.gm/IDAV (dBΩ)f−3dB (GHz)IRN (A/√Hz)Required value> 40≈ 28< 20 pSimulated (TIA)47.628.415.5 pgm/ID methodology50.6146.97.9 p Comparison of target specifications, simulation results, and gm/ID-based estimates for the modified cascode inverter-based TIA. Simulation results and layout implementation Modified TIA The simulated frequency response of the modified cascode inverter-based TIA is shown in Fig. 10. The proposed design achieves a transimpedance gain of 47.6 dBΩ, a − 3 dB bandwidth of 28.4 GHz, and a total power consumption of 183 µW. These results are obtained by combining resistive feedback, diode-connected devices, and active-inductor techniques. Fig. 10Simulated frequency response of the modified cascode inverter-based TIA. Simulated frequency response of the modified cascode inverter-based TIA. To provide a more accurate assessment of the small-signal behavior of the transistors used in the TIA, the drain current (ID) and equivalent drain-source resistance (rds) values are listed in Table 3. These parameters strongly influence the gain, bandwidth, power dissipation, and, consequently, the overall frequency response of the circuit. Table 3Drain current and equivalent drain–source resistance of the TIA transistors.CNT1CNT2CNT3CNT4CNT5CNT6CNT7CNT8ID474 nA474 nA502 nA502 nA0.21 pA203 µA3.15 pA203 µArds224 KΩ659 KΩ643 KΩ315 KΩ339 Ω2 KΩ600 Ω0.2 Ω Table 3Drain current and equivalent drain–source resistance of the TIA transistors.CNT1CNT2CNT3CNT4CNT5CNT6CNT7CNT8ID474 nA474 nA502 nA502 nA0.21 pA203 µA3.15 pA203 µArds224 KΩ659 KΩ643 KΩ315 KΩ339 Ω2 KΩ600 Ω0.2 Ω Drain current and equivalent drain–source resistance of the TIA transistors. Figure 11 shows the input-referred and output noise characteristics of the modified TIA. The input-referred noise current is approximately 15.5 pA/√Hz at low frequencies and 7.2 pA/√Hz at the − 3 dB frequency. Fig. 11(a) Input-referred noise and (b) output noise of the modified cascode inverter-based TIA. (a) Input-referred noise and (b) output noise of the modified cascode inverter-based TIA. Eye-diagram simulations were carried out using a 27–1 non-return-to-zero (NRZ) pseudo random binary sequence (PRBS) at 40 Gbps, with input currents of 10 µA and 50 µA and rise/fall times of 15 ps. For the 10 µA input current, the vertical eye opening (reported as ISI in the original data) is 2.19 mV and the horizontal eye opening (jitter) is 22.95 ps. For the 50 µA input current, the ISI increases to 11.33 mV, while the jitter remains nearly unchanged at 23.44 ps, as shown in Fig. 12. These results indicate stable eye openings over the examined input-amplitude range. Fig. 12Eye diagrams of the modified cascode inverter-based TIA for (a) 10 µA and (b) 50 µA input currents. Eye diagrams of the modified cascode inverter-based TIA for (a) 10 µA and (b) 50 µA input currents. Monte Carlo simulations were performed to evaluate the effect of ± 10% threshold-voltage (Vth) variations modeled as Gaussian-distributed gate-connected voltage sources41. A total of 200 simulation runs were conducted. The resulting variations in gain and bandwidth are presented in Fig. 13 through the frequency-response spread and the corresponding histograms. Fig. 13Monte Carlo simulation results for the modified cascode inverter-based TIA: (a) frequency response and histograms of (b) gain and (c) bandwidth under ± 10% Vth variation. Monte Carlo simulation results for the modified cascode inverter-based TIA: (a) frequency response and histograms of (b) gain and (c) bandwidth under ± 10% Vth variation. For layout implementation, the TIA employs CNTs with chirality (19,0) and a diameter of 1.5 nm. The CNT spacing is set to 20 nm, with an edge clearance of 10 nm. The effective gate width is defined as Wgate = N × (20 nm + 1.5 nm), where N denotes the number of CNTs per device. The gate length is 32 nm, corresponding to a normalized channel length of L = 2λ with λ = 16 nm. Table 4 lists the number of CNTs and the corresponding λ-based gate widths. Table 4Number of CNTs and λ-based gate widths for transistors in the modified cascode inverter-based TIA.CNT1CNT2CNT3CNT4CNT5CNT6CNT7CNT8 N 111111180480W (λ-based)21.521.521.521.52386.51720861720 Table 4Number of CNTs and λ-based gate widths for transistors in the modified cascode inverter-based TIA.CNT1CNT2CNT3CNT4CNT5CNT6CNT7CNT8 N 111111180480W (λ-based)21.521.521.521.52386.51720861720 Number of CNTs and λ-based gate widths for transistors in the modified cascode inverter-based TIA. The TIA layout occupies an area of 190λ × 81λ, corresponding to 3040 nm × 1296 nm, as shown in Fig. 14. Fig. 14Layout of the modified cascode inverter-based TIA. Layout of the modified cascode inverter-based TIA. LA stage The simulated frequency response of the LA stage with active-inductive loads is shown in Fig. 15. The LA provides a voltage gain of 10.3 dB, a − 3 dB bandwidth of 32.0 GHz, and a total power consumption of 382 µW. Fig. 15Frequency response of the LA with active-inductive loads. Frequency response of the LA with active-inductive loads. To further characterize the small-signal behavior of the LA, the drain current (ID) and equivalent drain-source resistance (rds) of its transistors are summarized in Table 5. Table 5Drain current and equivalent drain–source resistance of the LA transistors.CNT0CNT1CNT2CNT3CNT4CNT5CNT6ID424 µA212 µA212 µA212 µA2 pA212 µA2 pArds0.3 KΩ1.2 KΩ1.2 KΩ2.4 KΩ889 GΩ2.4 KΩ889 GΩ Table 5Drain current and equivalent drain–source resistance of the LA transistors.CNT0CNT1CNT2CNT3CNT4CNT5CNT6ID424 µA212 µA212 µA212 µA2 pA212 µA2 pArds0.3 KΩ1.2 KΩ1.2 KΩ2.4 KΩ889 GΩ2.4 KΩ889 GΩ Drain current and equivalent drain–source resistance of the LA transistors. Each transistor uses a normalized channel length of 2λ. Table 6 summarizes the number of CNTs and the corresponding λ-based gate widths. Table 6Number of CNTs and λ-based gate widths for the transistors in the LA.CNT0CNT1CNT2CNT3CNT4CNT5CNT6 N 205385385404404W (λ-based)275.46517.34517.3453.755.37553.755.375 Table 6Number of CNTs and λ-based gate widths for the transistors in the LA.CNT0CNT1CNT2CNT3CNT4CNT5CNT6 N 205385385404404W (λ-based)275.46517.34517.3453.755.37553.755.375 Number of CNTs and λ-based gate widths for the transistors in the LA. The LA layout occupies an area of 573λ × 76λ, equivalent to 9168 nm × 1216 nm, as shown in Fig. 16. Fig. 16Layout of the LA. Layout of the LA. Optical receiver system The complete low-noise optical receiver operating at 40 Gbps achieves a transimpedance gain of 77.8 dBΩ, a − 3 dB bandwidth of 29.5 GHz, and a total power consumption of 1377 µW. The input-referred noise current is 13.74 pA/√Hz. The corresponding frequency response is shown in Fig. 17. Fig. 17Frequency response of the complete optical receiver system. Frequency response of the complete optical receiver system. The overall optical receiver layout occupies an area of 9760 nm × 5040 nm, as shown in Fig. 18. Fig. 18Layout of the complete optical receiver system. Layout of the complete optical receiver system. Bias voltage generation circuits The use of multiple external bias voltages can increase the design and control complexity of the optical receiver. To address this issue, the proposed architecture employs two dedicated bias-voltage generation circuits to provide the two required bias voltages for the TIA and the single required bias voltage for the LA. These circuits generate the desired bias levels through voltage division, and their topologies are shown in Fig. 19. Fig. 19Bias-voltage generation circuits for (a) the TIA and (b) the LA. Bias-voltage generation circuits for (a) the TIA and (b) the LA. To obtain the target bias voltages of + 0.15000 V, − 0.15000 V, and − 0.04000 V, the CNT pitch is adjusted as an additional design parameter. Although the nominal global pitch is 20 nm, it is reduced to 13.5 nm in the TIA bias circuit and to 10.8 nm in the LA bias circuit. The power consumption of the TIA and LA bias circuits is 0.44 µW and 7.74 µW, respectively, resulting in a total bias-generation overhead of only 8.18 µW for the optical receiver. 40 Gbps performance comparison Table 7 summarizes the simulated performance of the proposed modified cascode inverter-based TIA, the LA, and the overall optical receiver (OR), and compares them with previously reported 40 Gbps designs. It should be emphasized that the results of the proposed CNTFET-based receiver are obtained exclusively from circuit-level simulations using the adopted 32 nm CNTFET device model and do not include post-layout parasitic extraction (PEX). Therefore, the comparison in Table 7 is intended only as a design-level benchmarking reference and should not be interpreted as a direct comparison with experimentally measured silicon prototypes or post-layout verified implementations. To facilitate quantitative performance evaluation, two figures of merit (FoMs) are defined as40\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:FOM1=\frac{Gain\:\left(dB\varOmega\:\right)\times\:Bandwidth\:\left(GHz\right)}{Power\:Consumption\:\left(mW\right)\times\:Input\:Referred\:Noise\:\left(pA/\sqrt{Hz}\right)}$$\end{document}41\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:FOM2=\frac{Gain\:\left(dB\varOmega\:\right)\times\:Bandwidth\:\left(GHz\right)\times\:{C}_{PD}\:\left(fF\right)}{Power\:Consumption\:\left(mW\right)\times\:Input\:Referred\:Noise\:\left(pA/\sqrt{Hz}\right)}$$\end{document} Both FoMs indicate that the proposed modified cascode inverter-based TIA and low-noise optical receiver achieve favorable performance relative to previously reported designs. The observed improvements are primarily attributed to the selected circuit topology, low-voltage operation, and the efficient transport characteristics of CNTFET devices. The photodiode capacitance, CPD, primarily determines the input pole of the TIA and therefore directly affects the receiver frequency response and bandwidth. In contrast, the DC or low-frequency transimpedance gain is mainly governed by the transconductance of the driver transistors and the effective output resistance. Accordingly, the reported reduction in power consumption should not be interpreted as a direct consequence of a smaller CPD. Rather, it mainly stems from the intrinsic device-level advantages of the CNTFET-based implementation, including higher achievable transconductance at lower bias current, reduced supply-voltage operation, and improved electrostatic gate control. In this work, the photodiode capacitance is assumed to be 10 fF to model an intrinsic, best-case device-level condition. This choice is consistent with the nanoscale CNTFET technology considered here, in which the intrinsic and parasitic capacitances of circuit nodes are typically on the order of a few femtofarads. Therefore, a value of 10 fF is adopted for both the photodiode capacitance and the capacitive load as the dominant effective capacitance at the input and output nodes. Since this value is approximately one order of magnitude larger than the equivalent parasitic capacitances of CNTFET-based nodes, it provides a reasonable basis for evaluating the intrinsic high-frequency performance of the proposed circuit within the scope of the present analysis. Accordingly, the selected 10 fF value should be interpreted as an idealized device-level assumption for assessing the intrinsic and high-speed performance of the proposed receiver, rather than as a complete representation of a practical optical receiver including all interconnect and packaging parasitics. Under this low-capacitance loading condition, the chosen value may therefore be regarded as physically plausible for evaluating the proposed design. Since a larger CPD may degrade the bandwidth, input-referred noise, and transient response, the comparison presented in Table 7 should be understood as a design-level benchmark rather than a fully normalized comparison under identical photodiode-capacitance conditions. Table 7Performance comparison of the modified cascode inverter-based TIA, LA, and optical receiver system with previously reported 40 Gbps designs.This work 42 43 44 45 46 47 48 49 50 Year2025201020102012201220152015201720212022Process Technology32 nmCNTFET0.13 μmMOSFET65 nmMOSFET45 nmSOICMOS65 nmMOSFET65 nmMOSFET0.13 μmBiCMOS0.13 μmBiCMOS0.18 μmBiCMOS90 nmMOSFETSupply voltage± 0.45 V1.5 V1.2 V1 V1.2 V1.2 V2.3 V2.3 V3.3 V1 VData rate Optical Receiver (OR)40 Gbps40 Gbps40 Gbps40 Gbps40 Gbps40 Gbps40 Gbps40 Gbps40 Gbps40 GbpsGain (TIA)47.6 dBΩ50 dBΩ46.7dBΩ55 dBΩ54 dB50 dBΩ13 dB―39.4 dBΩ40 dBΩBandwidth (TIA)28.4 GHz29 GHz21.6 GHz30 GHz25 GHz33.1 GHz43 GHz―32.9 GHz25.8 GHzPower consumption (TIA)183 µW45.7 mW39.9 mW9 mW――8.5 mW―77 mW29 mWInput referred noise TIA (pA/√Hz)15.551.83020.47――20―3725Area (TIA)3.93 µm20.4 mm27.5 m20.29 mm2――――0.45mm20.16 mm2FOM1476.580.61254.1008.95――――0.45491.42FOM24765.861.25820.09985.18――――――Gain (LA)10.3 dB―――10 dB―7 dB7 dB――Bandwidth (LA)32 GHz―――52.8 GHz―――――Power consumption (LA)382 µW―――――6.5 mW6.5 mW――Area (LA)11.14µm2―――――――――Gain (OR)77.8 dBΩ―――92 dBΩ79 dBΩ89.5 dBΩ80 dBΩ――Bandwidth (OR)29.5 GHz―――35 GHz29.6 GHz――――Power consumption (OR)1377 µW―――168 mW12.4 mW77 mW77 mW――Input referred noise OR (pA/√Hz)13.74―――1453.47――――Area (OR)49.19 µm2―――0.825 mm2―0.52 mm20.52 mm2――FOM1122.01―――6.10 µ3.5268――――FOM21220.1―――36.24 µ352.68――――CPD10 fF50 fF200 fF80 fF60 fF100 fF17 fF17 fF――Load10 fF―50 Ω50 Ω―――50 Ω50 Ω50 Ω Table 7Performance comparison of the modified cascode inverter-based TIA, LA, and optical receiver system with previously reported 40 Gbps designs.This work 42 43 44 45 46 47 48 49 50 Year2025201020102012201220152015201720212022Process Technology32 nmCNTFET0.13 μmMOSFET65 nmMOSFET45 nmSOICMOS65 nmMOSFET65 nmMOSFET0.13 μmBiCMOS0.13 μmBiCMOS0.18 μmBiCMOS90 nmMOSFETSupply voltage± 0.45 V1.5 V1.2 V1 V1.2 V1.2 V2.3 V2.3 V3.3 V1 VData rate Optical Receiver (OR)40 Gbps40 Gbps40 Gbps40 Gbps40 Gbps40 Gbps40 Gbps40 Gbps40 Gbps40 GbpsGain (TIA)47.6 dBΩ50 dBΩ46.7dBΩ55 dBΩ54 dB50 dBΩ13 dB―39.4 dBΩ40 dBΩBandwidth (TIA)28.4 GHz29 GHz21.6 GHz30 GHz25 GHz33.1 GHz43 GHz―32.9 GHz25.8 GHzPower consumption (TIA)183 µW45.7 mW39.9 mW9 mW――8.5 mW―77 mW29 mWInput referred noise TIA (pA/√Hz)15.551.83020.47――20―3725Area (TIA)3.93 µm20.4 mm27.5 m20.29 mm2――――0.45mm20.16 mm2FOM1476.580.61254.1008.95――――0.45491.42FOM24765.861.25820.09985.18――――――Gain (LA)10.3 dB―――10 dB―7 dB7 dB――Bandwidth (LA)32 GHz―――52.8 GHz―――――Power consumption (LA)382 µW―――――6.5 mW6.5 mW――Area (LA)11.14µm2―――――――――Gain (OR)77.8 dBΩ―――92 dBΩ79 dBΩ89.5 dBΩ80 dBΩ――Bandwidth (OR)29.5 GHz―――35 GHz29.6 GHz――――Power consumption (OR)1377 µW―――168 mW12.4 mW77 mW77 mW――Input referred noise OR (pA/√Hz)13.74―――1453.47――――Area (OR)49.19 µm2―――0.825 mm2―0.52 mm20.52 mm2――FOM1122.01―――6.10 µ3.5268――――FOM21220.1―――36.24 µ352.68――――CPD10 fF50 fF200 fF80 fF60 fF100 fF17 fF17 fF――Load10 fF―50 Ω50 Ω―――50 Ω50 Ω50 Ω Performance comparison of the modified cascode inverter-based TIA, LA, and optical receiver system with previously reported 40 Gbps designs. 32 nm CNTFET 0.13 μm MOSFET 65 nm MOSFET 45 nm SOICMOS 65 nm MOSFET 65 nm MOSFET 0.13 μm BiCMOS 0.13 μm BiCMOS 0.18 μm BiCMOS 90 nm MOSFET Conclusion This study presented the design and simulation of a 40 Gbps optical receiver implemented in 32 nm CNTFET technology. The receiver architecture consists of two identical modified cascode inverter-based TIAs followed by three cascaded differential LA stages. To improve high-frequency performance, the TIA incorporates resistor-feedback, diode-connected, and active-inductor techniques. All circuit blocks operate under a ± 0.45 V supply. The modified TIA achieves a transimpedance gain of 47.6 dBΩ, a − 3 dB bandwidth of 28.4 GHz, an input-referred noise current of 15.5 pA/√Hz, and a power consumption of 183 µW within an area of 3.93 μm². In addition, the analytical design based on the gm/ID methodology shows good agreement with the simulation results. The LA stage provides a voltage gain of 10.3 dB, a − 3 dB bandwidth of 32 GHz, and a power consumption of 382 µW, while occupying an area of 11.14 μm². At the system level, the complete optical receiver achieves a transimpedance gain of 77.8 dBΩ, a − 3 dB bandwidth of 29.5 GHz, and an input-referred noise current of 13.74 pA/√Hz, with a total power consumption of 1377 µW and an area of 49.19 μm². Overall, the simulation results, supported by analytical validation, indicate that the proposed design achieves a favorable trade-off between performance and power consumption. These characteristics suggest that the receiver is a suitable candidate for integration into next-generation energy-efficient optical communication systems. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2