Introduction

As the advantages of scaling silicon field-effect transistors (FETs) diminish1, transistors made from nanomaterials have gained significant attention due to their short-channel immunity and low power consumption2,3. Carbon nanotube FETs (CNTFETs), which use semiconducting carbon nanotubes (s-CNTs) as the channel material, have exhibited superior electrical performance and power efficiency compared to silicon FETs4,5,6,7. Notable advancements in material processing5,8,9, device fabrication4,10, system-level functionality11,12,13,14, and, most recently, successful compatibility with existing silicon manufacturing facilities15 have positioned CNTFETs on the brink of commercialization in semiconductor foundries16,17,18,19,20. However, the imperfect interfaces between CNTs and dielectrics remain a critical challenge. Induced gap states can facilitate carrier tunneling and trapping, resulting in undesirable power consumption marked by higher off-state current density (Ioff) and increased subthreshold swing (SS) values. The primary issue has been traced to organic molecules that arise during the post-synthesis separation of s-CNTs and throughout the device fabrication process. Consequently, there is an urgent need to explore techniques that can enhance CNTFET performance by addressing these organic contaminants, particularly in a manner that is compatible with foundry processes, featuring low thermal budgets, high throughput, and wafer-scale uniformity.

To enhance the quality of the CNT/dielectric interface by disrupting embedded organic materials, various forms of energy have been utilized to achieve the required bond dissociation energies (BDEs): (I) chemical energy21 (for decomposing polymers into smaller molecules), (II) thermal energy22 (for oxidative cracking of organics), and (III) electron-beam energy23,24 (for bond breaking through electron kinetics). However, these methods have limitations, including specificity to certain chemical bonds, excessively high thermal budgets, and potential changes to the doping state.

In addition, these approaches often introduce new interface states, resulting in only limited improvements – or even degradation – in the Ioff and SS of CNTFETs. For successful technology transfer to industry, an ideal technique should meet at least three key criteria: efficiency (high energy, deep penetration, precise controllability, and stability), performance (enhancements in electrical parameters related to interface quality, such as Ioff and SS), and manufacturability (low thermal budget, high throughput, scalability for wafer processing, and cost-effectiveness).

In this work, we propose that these requirements can be effectively fulfilled through the use of γ-ray irradiation, a technique that harnesses high-energy photons emitted from the decay of 60Co (half-life: 5.27 years)25. The energy of γ-rays, at 1.17 MeV and 1.33 MeV, is significantly higher than that of electron beams (approximately 10 keV) and the thermal energy associated with the annealing process (around 75 meV). Energy deposited through mechanisms such as Compton scattering typically results in permanent damage to chemical bonds, especially those with low BDEs (Supplementary Notes. 1-2, Supplementary Fig. 1). However, carbon nanotubes (CNTs), due to their nanoscale cross-sections, robust chemical bonds, and low backscattering capabilities, can endure prolonged exposure to a harsh γ-ray environment26,27,28. As a result, γ-ray irradiation can selectively disrupt organic compounds without harming the CNTs themselves. Furthermore, the power (energy per unit time) absorbed by a single CNT segment from γ-rays is substantially lower than that absorbed during thermal annealing (about 10−8 less) and electron-beam irradiation (approximately 10−13 less) (see “Methods”), indicating a high level of controllability with this approach. With improved interface quality, we achieved a reduction of 41 mV dec−1 in SS (from 175.9 to 134.9 mV dec−1) and a decrease of an order of magnitude in Ioff. The resulting on/off ratio of 105 reached the highest level among networked CNTFETs under saturated bias conditions. Moreover, this strategy offers high throughput, capable of processing over 8000 twelve-inch or 80,000 four-inch wafers per month (see Fig. 1a–c, Methods and Supplementary Fig. 2 for details). Combined with advantages such as a low thermal budget (operating at room temperature), cost-effectiveness (Supplementary Note. 3), high uniformity, and efficiency, γ-ray irradiation presents a foundry-compatible technique for CNTFET post-treatment, further paving the way for CNTFETs to become mainstream device technology.

Fig. 1: A foundry-compatible strategy for enhancing the performance of carbon nanotube field-effect transistors (CNTFETs) through γ-ray irradition.
Fig. 1: A foundry-compatible strategy for enhancing the performance of carbon nanotube field-effect transistors (CNTFETs) through γ-ray irradition.
Full size image

a Schematic diagram of γ-ray irradiation on wafer-scale CNTFETs. This room-temperature, high-throughput processing strategy meet the foundry requirements for efficiency, uniformity and cost-effectiveness. b Schematic diagram of the proposed quasi-gate-all-around (quasi-GAA) CNTFET, where the nanometer-thick CNT channel is modulated by two vertically connected gates. c Scanning electron microscopy (SEM) image of the deposited CNT network film. Scale bar: 100 nm. d Raman spectra of a single ultra-clean CNT before (0 Mrad(Si)) and after (100 Mrad(Si)) irradiation. The D peak remains negligible even after a total ionizing dose (TID) of 100 Mrad(Si), confirming the radiation tolerance of CNTs. The inset illustrates the process for obtaining an ultraclean CNT. e Raman spectra of conjugated PCz (poly[9-(1-octylonoyl)−9H-carbazole- 2,7-diyl]) molecules before (0 Mrad(Si)) and after (100 Mrad(Si)) irradiation. The inset displays the structural formula of PCz, with a characteristic wavenumber at 1622.4 cm−1. f Raman spectra of CNT network films irradiated at 0 Mrad(Si), 3 Mrad(Si), and 100 Mrad(Si), respectively. The typical peak of PCz is indicated by a dashed line. g X-ray photoelectron spectroscopy (XPS) spectra of CNT network films irradiated at 0 Mrad(Si), 3 Mrad(Si), and 100 Mrad(Si), respectively. The peak positions for sp2 and sp3 are marked with dashed lines. h Area ratio of sp2, sp3, C-O, and C = O of CNT network films irradiated at 0 Mrad(Si), 3 Mrad(Si), and 100 Mrad(Si), respectively.

Results

γ-ray irradiation on carbon nanotubes

To validate our hypothesis, γ-ray irradiation was initially performed on a single ultra-clean CNT segment (1-2 walls) that was detached from a multi-walled structure29,30. A custom-built transfer stage with two tungsten tips was used to precisely position the CNT onto a Si/SiO2 substrate (as shown in the inset of Fig. 1d, Supplementary Note. 4 and Supplementary Fig. 3). The Raman spectrum of the CNT segment exhibited an almost undetectable disordered signal at the D peak, confirming its pristine structure. The γ-ray irradiation was conducted at a dose rate of 300 rad(Si) s−1 at room temperature, a condition that was consistently maintained throughout all subsequent experiments (Supplementary Note. 5). As illustrated in Fig. 1d, the D peak remained negligible even after a total ionizing dose (TID) of 100 Mrad(Si) (equivalent to 92.6 hours, with 1 rad ≈ 0.01 J kg−1), clearly demonstrating the CNT’s immunity to γ-ray irradiation (Supplementary Note. 6).

The polymer-sorted CNT film was further subjected to irradiation and subsequently analyzed using Raman spectroscopy and X-ray photoelectron spectroscopy (XPS). S-CNTs were separated using conjugated PCz (poly[9-(1-octylonoyl)−9H-carbazole- 2,7-diyl]) molecules5,31,32,33 (as shown in the inset of Fig. 1e) and were then uniformly deposited onto a 4-inch Si/SiO2 substrate (Supplementary Note. 7). Results from scanning electron microscopy (SEM) (Fig. 1c, Supplementary Fig. 4, and Supplementary Note. 8) and atomic force microscopy (AFM) (Supplementary Fig. 5) collectively confirm the high density and uniformity of the networked films.

As demonstrated in Fig. 1f, the Raman peak at 1622.4 cm−1 exhibited a significant decrease following irradiation at a TID of 100 Mrad(Si) (Supplementary Note. 9). A detailed analysis indicated that this decline is likely due to the decomposition of the PCz polymer21 (Fig. 1e). Furthermore, the D/G intensity ratio was employed to quantify the level of disorder in the CNT films. The results show that the D/G ratio initially increased up to 3 Mrad(Si) before gradually decreasing, which correlated with improved material performance at 100 Mrad(Si). A similar trend was observed in aligned CNT films, where organic materials such as PCz and 2-butene-1,4-diol were also utilized for sorting and alignment (Supplementary Note. 10 and Supplementary Fig. 6). We hypothesize that the changes in the D/G ratio are connected to the breaking and reformation of chemical bonds within the organic materials, while the CNTs themselves remain relatively stable during irradiation (Fig. 1d).

To elucidate this, XPS spectra were obtained for CNT films irradiated at 0 Mrad(Si), 3 Mrad(Si), and 100 Mrad(Si), covering an energy range of 280– 292 eV (Fig. 1g and Supplementary Note. 11). As TID increased to 3 Mrad(Si), the proportion of chemical bonds with higher bond dissociation energies (CB-HBDEs), such as sp2 and C = O, showed a slight decrease, while the proportion of bonds with lower bond dissociation energies (CB-LBDEs), such as sp3 and C-O, increased slightly (Fig. 1h). This change may be attributed to the decomposition of certain bonds within the organic materials, such as those between single and double bonds (e.g., benzene rings). In contrast, at a TID of 100 Mrad(Si), the proportion of CB-HBDEs notably increased, while the proportion of CB-LBDEs significantly decreased. This trend corresponds well with the changes observed in the D/G ratio from the Raman results (Supplementary Fig. 7 and Supplementary Note 12), indicating that high-dose γ-ray irradiation preferentially decomposes CB-LBDEs, leading to the formation of CB-HBDEs as part of a structural reorganization process (Supplementary Note 13 and Supplementary Fig. 8).

To assess the electrical performance of s-CNT films after γ-ray irradiation, top-gate (TG) CNTFETs were fabricated using both the original (0 Mrad(Si)) and irradiated (100 Mrad(Si)) films, following a similar methodology (Supplementary Fig. 9). To eliminate the impact of trapped charges induced by irradiation in the Si/SiO2 substrate, a wet transfer process utilizing a sacrificial layer of polymethyl methacrylate (PMMA) was implemented to move the CNT films onto a new substrate (Supplementary Fig. 10). The results shown in supplementary Fig. 9 indicate a substantial reduction of 22.5% (332.9 mV dec−1) in SS, along with a slight decrease of 31.1% in Ioff (Vds = 2.0 V). It is worth noting that interface gap states were reintroduced during the device fabrication due to the use of organic materials, such as PMMA and photoresist. These findings preliminarily suggest that γ-ray irradiation can enhance device performance.

High-performance carbon nanotube transistors

A more suitable approach for industrialization would involve γ-ray post-treatment of the fabricated devices, rather than the CNT films themselves, which would help prevent the reintroduction of gap states during device processing. As shown in Fig. 1b, a quasi-gate-all-around (quasi-GAA) CNTFET is constructed on a 4-inch Si/SiO2 substrate using standard photolithography and lift-off techniques (refer to the photograph in Fig. 2a and the fabrication processes detailed in supplementary Fig. 11). The term – quasi-GAA – refers to a nanometer-thick CNT channel that is controlled by two vertically connected gates. This design improves device reliability and gate efficiency compared to a single-gate configuration, resulting in increased Ion and reduced SS. Unlike advanced GAA structures, the quasi-GAA configuration provides similar control effectiveness for devices with micro-scale channel lengths (Lch), while offering a less complex fabrication process. Moreover, while established P-type doping technologies are available for CNTFETs5,14,15,31,33,34,35,36, the development of N-type devices continues to pose significant challenges20,32,37,38,39,40. For example, low-work-function metals can create ohmic N-channel contacts, but their instability in air limits their use in irradiated environments. In this study, electrostatic doping through dielectric layers was employed, allowing for the tunneling injection of electrons from a palladium (Pd) electrode into the CNT channel (Supplementary Note. 14 and Supplementary Fig. 12). The channel length and width of the fabricated CNTFETs are 4 μm and 28 μm, respectively. Ti/Pd/Au (0.5/20/20 nm) stacks were utilized as the source/drain electrodes. The HfOx/HfAlOx layers, which were deposited using atomic layer deposition (ALD) with a target thickness of 10 nm, served as both the dielectric and doping layers (Supplementary Notes. 15, 16). Cross-sectional transmission electron microscopy (TEM) imaging paired with the energy-dispersive X-ray spectroscopy (EDS) mapping, conducted both before and after irradiation, collectively confirms the radiation robustness of the fabricated CNTFETs (Fig. 2b–d, Supplementary Fig. 13 and Supplementary Notes. 17, 18).

Fig. 2: High-performance quasi-GAA CNTFETs.
Fig. 2: High-performance quasi-GAA CNTFETs.
Full size image

a Photograph of the as-fabricated CNTFETs on a 4-inch Si/SiO2 substrate. b Cross-sectional high-resolution transmission electron microscope (TEM) image of the as-fabricated CNTFET. Scale bar: 10 nm. c Cross-sectional energy dispersive X-ray spectroscopy (EDS) mapping of the as-fabricated CNTFET, see Supplementary Note. 18 for more details. Scale bar: 10 nm. d EDS line analysis of the following elements: Gold (Au), Titanium (Ti), Hafnium (Hf), Aluminum (Al), and Oxygen (O). e Typical transfer curves of the CNTFET at drain-to-source voltage (Vds) of 2 V and 50 mV, respectively. The subthreshold swing (SS) of the device is 102.0 mV dec−1 at low bias (50 mV) and 173.7 mV dec−1 at saturated bias (2 V). Inset: False-color SEM image of the representative CNTFET, with a channel length (Lch) and width of 4 μm and 28 μm, respectively. Scale bar: 5 μm. f Typical output curves of the CNTFET with gate-to-source voltage (Vgs) varying from 2 V to 0 V in step of − 0.2 V. g Benchmarking of on-state current (Ion) versus SS of our results with reported N-type networked CNTFETs with micrometer Lch. Vds is labeled in the figure. Inset plots the experimental results of Ion versus TID.

The fabricated CNTFETs were characterized at room temperature and under ambient conditions. The typical transfer curves demonstrated a strong current density at Vgs = 2 V, confirming that electrostatic doping successfully converted the initial CNTs to N-type polarity (Fig. 2e). The output characteristics revealed promising current control and saturation (Fig. 2f). The linear relationship of Ids-Vds at low bias indicated that tunneling behaviors were not constrained by the sharp Schottky barrier in the contact region. This device achieved an impressive Ion of 9.82 μA μm−1 at a relatively low operating voltage (Vdd) of 2 V (Vgs = Vds = Vdd, Supplementary Note 19), significantly surpassing the values reported in the literature (which range from approximately 0.01-1 μA μm−1)20,32,33,34,39,41,42,43. Remarkably, the device achieved low SS values of 102.0 mV dec−1 at low bias (50 mV) and 173.7 mV dec−1 at saturated bias (2 V), despite having a relatively thick gate oxide stack (~ 10 nm). These performance enhancements exceeded those of N-type network-based CNTFETs. For comparison, the Ion versus SS values from the literature and those obtained in this study, focusing on micrometer-scale Lch, are plotted in Fig. 2g. This comparison highlights that our devices provide superior Ion while maintaining low SS, primarily due to the excellent gate efficiency of our quasi-GAA architecture (Fig. 2g, Supplementary Fig. 14, and Supplementary Note. 20) and the optimized fabrication processes that ensure a relatively clean interface.

γ-ray irradiation on carbon nanotube transistors

Subsequently, γ-ray irradiation was performed on the CNTFETs under unbiased conditions. After exposure to a specific dose, controlled by the irradiation time, tests were conducted to evaluate any changes in their electrical performance. A total of 25 devices were selected for statistical analysis. The transfer curves of these devices prior to irradiation exhibited minimal performance variability, indicating the suitability of the quasi-GAA devices for large-scale foundry production (Fig. 3a). Typically, irradiation-induced trapped charges can impair the performance of CNTFETs, particularly by causing shifts in Vth. However, for the proposed quasi-GAA architecture, trapped charges from the substrate and the surrounding environment were effectively shielded by the bottom and top gate electrodes, respectively. As a result, even after irradiation doses of up to 100 Mrad(Si), the fluctuation in Vth remained negligible (within 10% of Vdd), underscoring the robustness of the quasi-GAA architecture in radiation environments. In addition, the gate-to-source current (Igs) remained at an ultralow level (~3 pA) throughout the entire irradiation process, indicating that the dielectric layers were not compromised by the γ rays (Supplementary Note 21, and Supplementary Figs. 15, 16). Figure 3b illustrates the transfer curves of a typical device after exposure to doses of 0 Mrad(Si), 3 Mrad(Si), and 100 Mrad(Si). At 3 Mrad(Si), some performance degradation was observed, characterized by a 2.4-fold increase in Ioff and a 28.5 mV dec−1 rise in SS. However, surprisingly, after irradiation at 100 Mrad(Si), the device performance improved: Ion increased by 54.9% to 14.33 μA μm−1, Ioff decreased by more than an order of magnitude to 157.7 pA μm−1, and SS reduced by 57.0 mV dec−1 to 144.7 mV dec−1.

Fig. 3: γ-ray irradiation effects on CNTFETs.
Fig. 3: γ-ray irradiation effects on CNTFETs.
Full size image

a Transfer curves of 25 devices used for subsequent irradiation tests. The inset displays the median values of typical electrical parameters. b Transfer curves of a typical CNTFET irradiated at 0 Mrad(Si), 3 Mrad(Si), and 100 Mrad(Si), respectively. Solid lines represent logarithmic coordinates, and dashed lines represent linear coordinates. c Band diagrams of the CNTFETs irradiated at 3 Mrad(Si) (worst condition). d Band diagrams of the CNTFETs irradiated at 100 Mrad(Si) (best condition). e Performance variation of CNTFETs and CNT films within the TID range of 0 to 100 Mrad(Si). As the TID increases, device performance initially deteriorates and then gradually improves, with an inflection point at 3 Mrad(Si). Within 100 Mrad(Si), the threshold voltage (Vth) and gate-to-source current (Igs) experience negligible changes. The box plots show the median (center line), interquartile range (box boundaries), and minimum/maximum values (whiskers).

These variations were further supported by statistical analysis. It was observed that the trends of Ion and Ioff consistently displayed opposite behaviors, with their changes being markedly different. This suggests that the counterintuitive behavior of the devices was primarily influenced by channel properties rather than contact resistance (Supplementary Notes. 22, 23 and Supplementary Figs. 17, 18). This conclusion is reinforced by the correlation between the materials’ D/G ratio and the devices’ performance. Specifically, an increasing proportion of CB-LBDEs, such as sp3 and C-O, was consistently associated with a higher D/G ratio and deteriorating device performance, evidenced by decreased Ion, increased Ioff, and higher SS values. In contrast, an increasing presence of CB-HBDEs, such as sp2 and C = O, correlated with a lower D/G ratio and enhanced device performance (Supplementary Note 24 and Supplementary Fig. 17). We attribute this behavior to gap states within the bandgap of CNTs (Fig. 3c, d). These gap states, typically resulting from chemical doping interactions involving adjacent CB-LBDEs, can facilitate the electron tunneling process from the Pd source electrode to the CNT channel, leading to increased Ioff. In addition, the density of interface states can also enhance SS values (Supplementary Note 25). However, γ-ray irradiation tends to convert these CB-LBDEs into CB-HBDEs, which are often linked to the formation of delocalized π electrons, introducing electron doping in the conduction band of s-CNTs or deep within the bandgap. As the shallow gap states are reduced, both Ioff and SS can significantly decrease, while the improvement in Ion is less pronounced due to the relatively low content of organics, achieved through optimized material selection and device fabrication processes (Fig. 3e, Supplementary Fig. 18, Supplementary Note. 19 and Supplementary Note 26).

Reliable γ-ray irradiation strategy

The effectiveness of γ-ray irradiation in enhancing electrical performance was further confirmed through statistical measurements. As shown in Fig. 4a, b, a relatively low device-to-device variation was noted before and after irradiation across 100 devices fabricated on a Si/SiO2 substrate, indicating the viability of the proposed quasi-GAA architecture for large-scale manufacturing and γ-ray treatment. Furthermore, Vth of the irradiated devices remained nearly unchanged (approximately 1.4% of Vdd), while SS decreased by 41 mV dec−1 to 134.9 mV dec−1, and Ioff was reduced by an order of magnitude to 112.2 pA μm−1 (Supplementary Fig. 19). These improvements align with the previously mentioned results, underscoring that γ-ray irradiation serves as a robust and effective method for enhancing the performance of CNTFETs. Notably, γ-ray irradiation on non-silicon substrates (photosensitive polyimide, PSPI) and CNTFETs with 600 nm channel lengths – including both network-based and array-based architectures – can also achieve similar enhancements, further demonstrating the reliability of this technique (Fig. 4c, d, Supplementary Notes 27–30, and Supplementary Figs. 20–23). Importantly, since irradiation causes permanent damage to polymers, the performance improvements do not degrade over time (Supplementary Note 31 and Supplementary Fig. 24).

Fig. 4: Reliable γ-ray irradiation strategy.
Fig. 4: Reliable γ-ray irradiation strategy.
Full size image

a Transfer curves of 100 devices, fabricated on a Si/SiO2 substrate, before (0 Mrad(Si)) and after (100 Mrad(Si)) irradiation. b Statistical analysis of the Vth, SS, and Ioff of devices in (a). c Transfer curves of 100 devices, fabricated on a Si/SiO2/PSPI substrate, before (0 Mrad(Si)) and after (100 Mrad(Si)) irradiation. d Statistical analysis of the Vth, SS, and Ioff of devices in (c). e Benchmarking of the on/off ratio versus Ioff of our results at Vds = 2 V with reported CNTFETs with micrometer-scale Lch. Hollow and solid patterns represent N-type32,33,39,40,43 and P-type14,33,49 CNTFETs, respectively. Lch (μm) is labeled in the figure. The low-power target for transistors at 100 pA μm-1 is indicated by a dashed line. The inset plots the experimental results of the on/off ratio and TID. f Benchmarking of TID tolerance versus \(\Delta {V}_{{{{\rm{th}}}}}/{V}_{{{{\rm{dd}}}}}\) of our results with reported radiation-hardened transistors. The hollow pattern represents silicon transistors51,52,53,54,55, while the solid pattern represents CNTFETs26,27,28,56,57,58,59,60. Dose rate (rad(Si) s−1) is labeled in the figure, and NA indicates the dose rate is not available in those references.

Discussion

Under ultra-small bias conditions, CNTFETs are often characterized by low Ioff and high on/off ratios5,20,26,31,32,34,44,45,46,47. However, assessing these parameters at operational voltages (Vdd) is more pertinent for practical circuit applications48, such as complementary metal oxide-semiconductor (CMOS) technology and ring oscillators (Supplementary Fig. 25). Although quasi-GAA CNTFETs have shown an order of magnitude improvement in both Ioff and on/off ratio compared to as-fabricated TG devices, their performance remains moderate compared to other reported CNTFETs with micrometer-scale channel lengths14,15,32,33,39,40,43,49. Further enhancements are hindered by the presence of organic-induced gap states. With the proposed γ-ray treatment, however, the performance of quasi-GAA CNTFETs in terms of Ioff and on/off ratio is improved by another order of magnitude, achieving the highest levels for these metrics (Fig. 4e). Typically, a dose of 100 Mrad(Si) can reduce Ioff to the low-power target15 of 100 pA μm−1 due to the reduction in gap states.

Another notable finding of this study is the exceptional radiation tolerance demonstrated by our quasi-GAA CNTFETs, capable of withstanding approximately 100 Mrad(Si). This property shows considerable potential for deployment in high-radiation scenarios like deep-space missions, nuclear reactor operations, and radiation therapy procedures50. Traditional silicon-based transistors, even those designed with radiation-hardened processes, generally fail to endure TID levels of 1 Mrad(Si)51,52,53,54,55. In contrast, CNT channels typically exhibit reported TID tolerances ranging from 1 to 10 Mrad(Si)26,27,28,56,57,58,59,60, thanks to their lower hit probabilities and higher BDEs. The proposed quasi-GAA architecture, with its nearly ideal electrostatic shielding, achieves a TID tolerance of 100 Mrad(Si) without compromising the Vth by more than 10% of the supply voltage (Vdd). This represents a two-order-of-magnitude improvement over silicon devices and a one-order-of-magnitude enhancement compared to previously reported CNTFETs (Fig. 4f). Despite these advancements, further research is necessary before integrating CNTFETs into practical irradiation environments, particularly to understand their behavior under operational conditions.

We have developed a foundry-friendly method to enhance the performance of CNTFETs through γ-ray irradiation. This room-temperature processing technique is compatible with wafers of various sizes, ranging from 4 to 12 inches, facilitating large-scale production of CNTFETs while meeting foundry requirements for high efficiency, uniformity, and cost-effectiveness. Characterization using Raman spectroscopy and X-ray photoelectron spectroscopy revealed that high-energy, low-beam γ rays do not damage the pristine structure of CNTs but instead induce a transformation in the chemical bonds of the organics surrounding them, shifting from low-energy to high-energy bonds. This transformation was further validated at the device level through the fabrication of quasi-GAA CNTFETs. The comprehensive performance metrics, including an Ion of 9.82 μA μm−1 and SS values of 102.0 mV dec−1 at a 50 mV bias and 173.7 mV dec−1 at a 2 V saturated bias, represent some of the best results reported in the literature for N-type networked CNTFETs with micrometer-scale channel lengths (Lch). Statistical analysis of γ-ray irradiation across 100 devices indicated that the SS decreased by 41 mV dec−1, reaching 134.9 mV dec−1. The Ioff was reduced by an order of magnitude to 112.2 pA μm−1, approaching the low-power transistor target of 100 pA μm−1. The on/off ratio (~105) achieved the highest level recorded among networked CNTFETs under saturated bias conditions (Supplementary Note 32). The performance enhancements resulting from γ-ray irradiation can be attributed to the reduction of shallow gap states due to an increase in the content of CB-HBDEs. The proposed quasi-GAA architecture exhibits a Vth shift of less than 10% of Vdd during a TID of 100 Mrad(Si), outperforming traditional silicon-based devices by at least two orders of magnitude and advanced radiation-hardened CNTFETs by one order of magnitude. In summary, our work presents an optimal, foundry-friendly strategy for enhancing the performance of CNTFETs through γ-ray treatment, broadening their application prospects in radiation-heavy environments.

Methods

Power estimation of different processing methods

This section will separately estimate the power (energy per unit time) of three processing methods on a single CNT segment: γ-ray irradiation, high-temperature annealing, and e-beam irradiation. Given that the diameter of a focused e-beam is approximately 10 nm (De-beam), we will consider a CNT segment with a length (L) of 10 nm and a diameter (d) of 1 nm for estimation. This simplification will not affect the qualitative conclusions.

In γ-ray irradiation, the dose rate, defined as the energy acting on unit mass per unit time, is commonly used to quantify the intensity of γ-rays. The mass (mCNT) of the CNT segment can be estimated by calculating the number of carbon atoms (Ncarbon) using the following relationship.

$${N}_{{{{\rm{carbon}}}}}=\frac{L}{{l}_{{{{\rm{CB}}}}}}\times \frac{\pi d}{{l}_{{{{\rm{CB}}}}}}=\frac{L\pi d}{{l}_{{{{\rm{CB}}}}}^{2}}$$
(1)

where lCB is the bond length of CNTs (lCB = 0.142 nm). Thus, the mass of the CNT segment is:

$${m}_{{{{\rm{CNT}}}}}={N}_{{{{\rm{c}}}}{{{\rm{arbon}}}}}\times {m}_{0}$$
(2)

here, m0 is the mass of a single carbon atom (\({m}_{0}=1.993\,\times {10}^{-26}{{{\rm{kg}}}}\)).

In our experiment, a dose rate of 300 rad(Si) s−1 is adopted. It should be emphasized that this value is calibrated using silicon materials. For CNTs, the corresponding conversion relationship is:

$${D}_{{{{\rm{CNT}}}}}={D}_{{{{\rm{Si}}}}}\times \left(\frac{{\left(\mu /\rho \right)}_{{{{\rm{CNT}}}}}\cdot {\rho }_{{{{\rm{CNT}}}}}}{{\left(\mu /\rho \right)}_{{{{\rm{Si}}}}}\cdot {\rho }_{{{{\rm{Si}}}}}}\right)$$
(3)

where \({D}_{{{{\rm{CNT}}}}({{{\rm{Si}}}})}\) is the dose rate calibrated by the CNT (silicon). The mass attenuation coefficient (MAC) of CNTs and silicon are estimated to be 0.0569 cm2 g−1 and 0.05688 cm2 g−1, respectively61. The densities of CNTs and silicon are 1.6 g cm−3 and 2.33 g cm−3, respectively. Consequently, the dose rate quantified by CNTs (\({D}_{{{{\rm{CNT}}}}}\)) is 206.1 rad(CNT) s−1 or 2.061 J kg−1 s−1.

Finally, the power acting on the single CNT segment from γ-rays (\({P}_{{{{\rm{\gamma }}}}-{{{\rm{ray}}}}}\)) is estimated to be \(6.4\times {10}^{-21}\,{{{\rm{W}}}}\) using the following formular:

$${P}_{{{{\rm{\gamma }}}}-{{{\rm{ray}}}}}={m}_{{{{\rm{CNT}}}}}\times {D}_{{{{\rm{CNT}}}}}$$
(4)

In high-temperature annealing, the annealing temperature is typically set to 600 °C (\(T=873.15\,{{{\rm{K}}}}\)), and the power of thermal radiation can be described using the Stefan-Boltzmann law (Supplementary Note 33). According to this law, the thermal radiation power (\({P}_{{{{\rm{thermal}}}}}\)) is given by:

$${P}_{{{{\rm{thermal}}}}}=\varepsilon \sigma {T}^{4}A$$
(5)

where \(\varepsilon\) is the emissivity of carbons, which is approximately 0.8162, \(\sigma\) is the Stefan-Boltzmann constant, \(A\) is the surface area of the CNT segment, calculated as \(A=\pi {dL}\).

By substituting the respective values, the power for high-temperature annealing is estimated to be \(8.39\times {10}^{-13}\,{{{\rm{W}}}}\).

In e-beam irradiation, assuming the e-beam energy (\({E}_{{{{\rm{e}}}}-{{{\rm{beam}}}}}\)) is 10 keV and the specimen current (\({I}_{{{{\rm{e}}}}-{{{\rm{beam}}}}}\)) is 100 pA, the number of electrons in the e-beam per second can be calculated using the following formular:

$${N}_{{{{\rm{e}}}}}=\frac{{I}_{{{{\rm{e}}}}-{{{\rm{beam}}}}}}{e}$$
(6)

where e is the elementary charge. The power acting on the single CNT segment from the e-beam can be calculated using:

$${P}_{{{{\rm{e}}}}}={\eta N}_{{{{\rm{e}}}}}\times {E}_{{{{\rm{e}}}}-{{{\rm{beam}}}}}$$
(7)

here, \(\eta\) is the action possibility, which has a minimum value of 0.128 determined by the area ratio of the CNT segment (\(\frac{{Ld}}{\pi {\left({D}_{{{{\rm{e}}}}-{{{\rm{beam}}}}}/2\right)}^{2}}\)).

Thus, the minimum power for e-beam irradiation is estimated to be \(1.27\times {10}^{-7}\,{{{\rm{W}}}}\).

Estimation of wafer processing throughput

As depicted in Supplementary Fig. 2, the γ-ray source (height: \(L\)) is positioned at the center of the irradiation environment. For a ring with a specific radius \({R}_{0}\), the number of wafers (diameter: \(D\)) that can be treated is given by:

$${N}_{0}=\left\lfloor L/D\right\rfloor \times \left\lfloor \left(2\pi {R}_{0}\right)/D\right\rfloor$$
(8)

where \(\left\lfloor x\right\rfloor\) donates rounding down the calculated value \(x\).

Assuming that wafers are arranged in rings with radii ranging from \(r\) to \(R\), and the spacing between adjacent rings is \(\Delta r\), the total number of wafers (\(N\)) that can be processed simultaneously in this irradiation setup is:

$$N={\sum }_{{R}_{0}=r}^{{R}_{0}=R}{N}_{0}$$
(9)

For objectivity, let’s consider \(r=0.5\,{{{\rm{m}}}}\), \(R=2.0\,{{{\rm{m}}}}\), \(\Delta r=0.1\,{{{\rm{m}}}}\), and \(l=1.0\,{{{\rm{m}}}}\). Then, for 12-inch wafers (\(D=300\,{{{\rm{mm}}}}\)), \(N={{\mathrm{1,230}}}\), while for 4-inch wafers (\(D=100\,{{{\rm{mm}}}}\)), \(N={{\mathrm{12,480}}}\). In our experiment, an irradiation dose of 100 Mrad(Si) required 92.6 hours. As an estimate, this setup could process ~ 8610 12-inch wafers or ~ 87,360 4-inch wafers.

Growth and manipulation of ultralong carbon nanotubes

Ultralong CNTs were grown on hollow Si/Si3N4 substrates using the chemical vapor deposition (CVD) method. Initially, a 0.2 nm iron catalyst was deposited on a Si/SiO2 substrate through e-beam evaporation. Subsequently, a super-aligned cross-stacked CNT film was placed on the surface of the substrate63. This film facilitated the detachment of the grown CNTs from the Si/SiO2 substrate. Under the influence of an airflow containing C2H4 (1.0 sccm), Ar (452 sccm), and H2 (216 sccm), the elongating CNTs were directed to drift and eventually settled on the pre-positioned hollow Si/Si3N4 substrate.

Ultralong suspended CNTs were then manipulated using two tungsten tips prepared via electrochemical etching. Initially, supported by a hollow Si/Si3N4 substrate, the CNTs were positioned above sulfur (S) vapor, allowing S particles to condense on their surface. This step was crucial for visualizing the nanoscale CNTs under an optical microscope. With the help of manipulators, the tungsten tips were used to touch and slide the end of a single CNT, causing it to detach from the substrate and adhere to the tip surface through van der Waals forces. Subsequently, a designed substrate was positioned beneath the tips, and the tips were maneuvered to place the CNT onto the specified substrate. Finally, the substrate was heated on a 170 °C hot plate for 2 min to remove the S particles from the CNT surface.

Preparation of carbon nanotube films

CNTs synthesized via arc-discharge using Ni/Y catalysts were sourced commercially from Carbon Solutions Inc. For selective extraction of semiconducting CNTs, the conjugated polymer PCz – synthesized through Suzuki polycondensation – served as a sorting agent. In this purification protocol, 100 mg of raw CNTs and 200 mg of PCz were dispersed in 100 mL of toluene. The suspension underwent 30 min ultrasonication with a top-tip dispergator (Sonics VC500), followed by centrifugation at 30,000 × g for 2 h (Allegra X-64R centrifuge) to isolate a semiconducting-enriched supernatant. Prior to film fabrication, substrates were cleaned in sequential solvent baths (toluene, acetone, and isopropanol). These substrates were then submerged in the purified CNT dispersion for 24 h, dried under a nitrogen stream, and annealed at 120 °C for 30 min on a hotplate.

Well-aligned CNT arrays were obtained using a dimension-limited self-alignment (DLSA) procedure. In detail, the high-purity semiconducting CNT solution was re-dispersed in 1,1,2-trichloroethane. The substrate was then vertically inserted into the CNT solution, followed by dropping a C4H8O2 solution. As the substrate was slowly extracted from the solutions, CNTs became confined to the 2D interface between the C4H8O2 and 1,1,2-trichloroethane solutions. Due to the strong affinity of C4H8O2 for SiO2, the CNTs assembled onto the substrate surface almost in parallel.

Transfer of carbon nanotube films

PMMA was utilized as a sacrificial layer to transfer the CNT film to an unirradiated Si/SiO2 substrate (Supplementary Fig. 10). Initially, a layer of PMMA was spin-coated onto the irradiated substrate at 3000 r min−1 for 1 min. The substrate was then immersed in electronic-grade ammonia water at room temperature. After the PMMA-coated CNT film detached from the substrate, it was rinsed twice with ultra-pure water. The PMMA film was subsequently lifted off using a clean, unirradiated substrate. Once the residual water evaporated naturally, the sample was baked on a hotplate at 90 °C for 5 min. Finally, the CNT film was transferred following sequential treatments with acetone, N-methyl-2-pyrrolidone (NMP), and ultra-pure water.

Characterizations of carbon nanotubes

Raman spectroscopy was conducted on the HORIBA LabRAM HR Evolution spectrometer with a laser wavelength of 532 nm and an 1800 mm−1 grating. The spectral data were calibrated using the silicon characteristic peak at 520.7 cm−1 on an undoped silicon specimen. Prior to Raman measurements on an individual CNT, its position was roughly determined using S vapor tagging. Subsequently, Raman line scanning was employed to collect spectroscopic data, with the data featuring the strongest signal being considered as the Raman spectrum of the CNT. For Raman analysis of PCz, PCz was dissolved in toluene and then spin-coated onto a Si/SiO2 substrate at 3000 r min−1 for 1 min.

XPS data were acquired using a Thermo Fisher Scientific ESCALAB Xi+ spectrometer with a best energy resolution of less than 0.43 eV. All spectral data were calibrated using the C1s peak at 284.8 eV. High-resolution TEM images were observed using an FEI Talos F200S G2 equipment. SEM was conducted using a Zeiss GENMINI SEM 500 setup. AFM was performed with a Bruker Dimension FastScan operating in tapping mode. Optica images were captured using a Nikon MM-400/L.

Device fabrication and measurements

Standard techniques were employed to fabricate all CNTFETs (Supplementary Fig. 11). For instance, the electrodes were fabricated using photolithography and lift-off processes. The vacuum level during e-beam evaporation was typically better than 10−4 Pa. The gate insulating layers were grown using the ALD at a temperature of 90 °C, with a specified thickness of 10 nm. To enhance nucleation sites on the CNT surfaces and thereby improve material growth quality, water vapor was introduced as the initial precursor.

For dual-gate devices, Ti/Au (1/40 nm) electrodes were initially patterned and evaporated onto a precleaned substrate. Subsequently, the HfAlOx bottom gate dielectric was grown using the ALD technique. This was followed by the deposition of the CNT network film. It should be emphasized that 10 nm of Y was deposited and subsequently removed using dilute hydrochloric acid to further clean the CNT films64. Next, the CNT channel was defined through an oxygen plasma process, and the source/drain electrodes (Ti/Pd/Au ~ 0.5/20/20 nm) were patterned. Following this, the HfOx top gate dielectric was grown using the ALD technique as well. Finally, the top gate electrode, composed of Ti/Au (1/40 nm), was patterned and evaporated.

For the PSPI substrate shown in Fig. 4c, a layer of PSPI photoresist (~500 nm thick) was spun onto a Si/SiO2 substrate and baked at 120 °C, 180 °C, 240 °C, and 300 °C for 30 min each to fully solidify the polymer. The fabrication processes for TG and BG devices were very similar to that of quasi-GAA devices. Notably, the TG device did not involve the preparation of the bottom-gate electrode and bottom-gate dielectric, while the BG devices did not require the preparation of the top-gate electrode. All electrical measurements were carried out at room temperature and atmospheric pressure, utilizing self-made programs based on the LabVIEW software.

The detailed extraction methods of Vth could be found in ref. 65. In our work, Vth was extracted from the gate voltage axis intercept of the \(\sqrt{{I}_{{{{\rm{ds}}}}}} \sim {V}_{{{{\rm{gs}}}}}\) characteristics linearly extrapolated at its maximum first derivative point.

Irradiation process

In this experiment, 60Co served as the irradiation source with a half-life of 5.27 years. Its decay produced non-radioactive 60Ni, accompanied by the emission of β particles and two types of γ rays with energies of 1.17 MeV and 1.33 MeV (\({}_{27}{}^{60}{{{\rm{Co}}}}\to {}_{28}{}^{60}{{{\rm{Ni}}}}+{}_{-1}{}^{0}{{{\rm{\beta }}}}+{}_{0}{}^{0}{{{\rm{\gamma }}}}\)). A stainless-steel casing was employed to shield the β particles, thereby ensuring a stable emission of γ rays. The cobalt rod was typically housed in a water tank and was only retrieved for the experiment when necessary.

The TID was directly proportional to the irradiation duration. In our experiment, the dose rate, which is the proportionality constant, was maintained at 300 rad(Si) s−1. To achieve a TID of 100 Mrad(Si), a total irradiation time of 92.6 h was required.