Junctionless Transistors: State-of-the-Art

Recent advances in semiconductor technology provide us with the resources to explore alternative methods for fabricating transistors with the goal of further reducing their sizes to increase transistor density and enhance performance. Conventional transistors use semiconductor junctions; they are formed by doping atoms on the silicon substrate that makes p-type and n-type regions. Decreasing the size of such transistors means that the junctions will get closer, which becomes very challenging when the size is reduced to the lower end of the nanometer scale due to the requirement of extremely high gradients in doping concentration. One of the most promising solutions to overcome this issue is realizing junctionless transistors. The first junctionless device was fabricated in 2010 and, since then, many other transistors of this kind (such as FinFET, Gate-All-Around, Thin Film) have been proposed and investigated. All of these semiconductor devices are characterized by junctionless structures, but they differ from each other when considering the influence of technological parameters on their performance. The aim of this review paper is to provide a simple but complete analysis of junctionless transistors, which have been proposed in the last decade. In this work, junctionless transistors are classified based on their geometrical structures, analytical model, and electrical characteristics. Finally, we used figure of merits, such as Ion/Io f f , DIBL, and SS, to highlight the advantages and disadvantages of each junctionless transistor category.


Classification
Junctionless transistors can be divided in two main categories: depletion-based and tunnel-based JLTs. In depletion-based devices, the amount of electrical current flowing through the device depends on the dimension of the depletion region, which is controlled by the applied gate voltage. In tunnel-based devices, the electrical current is governed by band-to-band tunneling (BTBT). JLT categories can be then classified based on the geometrical shape, material composition of the channel, as well as the gate structure. When a single gate on the top of the channel controls the current through the device, it is known as Single Gate junctionless transistor. If an additional gate is present below the channel, then it is called Double Gate junctionless transistor. If the transistor channel thickness is very thin (≤10 nm) and the channel material is not monocrystalline silicon, then it is known as Thin Film junctionless transistor. Gate-All-Around JLTs are characterized by a gate electrode that entirely surrounds the channel of the transistor; the channel can be either cylindrical or rectangular. When the transistor channel is a tube-shaped nanostructure, it is called junctionless Nanowire. If the electronic device is a fin-shaped transistor, it is called junctionless FinFET. This simple classification is coherent with the analyzed literature, whose distribution is shown in Figure 1. Although most of the reported junctionless transistors are inorganic electronic devices, organic solutions were also proposed [103][104][105][106][107][108]. In this work, we focus on inorganic junctionless transistors, where the following parameters as considered as the main figures of merit: the Drain-Induced Barrier Lowering (DIBL), which represents the drain voltage influence on the threshold voltage, defined as DIBL = |∆V th |/|∆V ds | [109]; the Subthreshold Swing (SS), which can be defined as the change in the gate voltage required to decrease the drain current by one decade (SS = dV gs /dlog(I d )) [110]; the I on /I o f f ratio, which is the ratio between the maximum available drain current, I on (V gs = V dd , V ds = V dd ), and the current in the off state,

The First Junctionless Transistor
In 2010, J. P. Colinge et al., fabricated the first junctionless transistor (Figure 2) in the form of a nanowire characterized by a silicon thickness of 10 nm and a channel length of 1 µm [2]. The process requires uniform and heavy doping (≥1 × 10 19 cm −3 ) of the nanowire in order to ensure the correct operation of the device [111]. The nanowire transistor was realized over a SOI (Silicon on Insulator) wafer and patterned by electron-beam lithography. Junctionless NWs do not work in inversion mode (IM) like conventional MOSFETs. The threshold voltage is defined as the gate voltage that fully depletes (OFF state) the device layer ( Figure 3c) [112,113]. Thus, the OFF state definition is more similar to that of accumulation mode (AM) devices ( Figure 3b); also from an analytical point of view, they can be modelled from AM device descriptions, since the current flowing in the device can be decomposed into two components: a bulk and an accumulation current [114,115]. When the gate voltage is greater (in absolute value) than the threshold voltage, partial depletion takes place creating a path in the substrate for the bulk current. When the gate voltage reaches the flat band voltage, the whole channel becomes conductive (ON state), and an accumulation current starts to form at the semiconductor/insulator interface [112]. The threshold and flat band voltages are of critical importance, since they determine the operating range of the device [2]. The threshold voltage can be extracted as the gate voltage at which the curve g m /I D (where g m is the transconductance) drops to half of its maximum value [116]. It increases (in absolute value) with increasing doping, while decreasing with increasing gate oxide thickness, nanowire width, and length [117]. An analysis of the doping concentration influence on the threshold voltage is also reported in [118]: by increasing the doping concentration from 1 × 10 14 cm −3 to 1 × 10 18 cm −3 , the threshold voltage decreases from 0.455 V to 0.37 V; this could be attributed to the fact that, for high doping concentration, more carriers are available for the same gate voltage. An interesting characteristic of the flat band voltage is the dependency on the temperature. A detailed analysis of this parameter is presented in [119,120], in which it is shown that the flat band voltage decreases as the temperature increases. For low temperature (4.2 K) and high doping concentration (1 × 10 19 cm −3 ), the thermal energy is not strong enough to ionize all of the dopants. This results in an incomplete ionization that causes the series resistance to increases ( Figure 4) and the drain current to decrease. As reported in [121], junctionless nanowire transistors present a reduction of short channel effects as compared to inversion mode devices, but they are highly sensitive to the series resistance. The simulation of junctionless NWs with channel length of 100 nm and doping concentration of 1 × 10 19 cm −3 provided an intrinsic source/drain resistance of 5.5 kΩ. By decreasing the doping to 5 × 10 18 cm −3 , the resistance reached a value of 11 kΩ. The performance of junctionless nanowire transistors can also be affected by current leakages that are associated to gate tunneling. The latter phenomenon is directly proportional to the length and width of the transistor (so to the gate surface area), and to the temperature [122,123].
Because the current in junctionless transistors flows far from the gate oxide/channel interface, the mobility degradation is minimized. This is one of the main advantages of depletion-based junctionless transistors with respect to the conventional ones [124]. Enhanced mobility in junctionless NWs was also attributed to the reduction of the scattered impurities that are caused by an overall smaller charge of ionized impurities [125]. Junctionless NWs were also investigated in terms of crystal orientations and material channel (germanium and silicon) and compared to inversion mode devices; the junctionless NW found to be less sensitive to short channel effects, presenting smaller SS, reduced DIBL, and higher I on /I o f f than inversion mode transistors. As compared to silicon JLTs, germanium junctionless NWs were observed to be slightly more sensitive to short channel effects, but more competitive from an electrostatic control point of view [126][127][128]. Figure 5 shows a schematic of a cylindrical junctionless Gate-All-Around transistor. The device, as the name suggests, is characterized by a channel entirely surrounded by the gate. The channel geometry determines the complexity of the equations that are needed to model the device behavior. In this case, the solutions to the Poisson equations are more complex, since cylindrical coordinates have to be introduced [26,31]. Additionally, GAAFETs with rectangular channels are reported, but they suffer from performance degradation due to corners effects [25]. An important parameter in the design of GAAFETs is the channel length. For a channel length reduction from 40 nm to 16 nm, the DIBL increases from 12 mV/V to 123 mV/V, while the SS increases from 62 mV/dec to 82 mV/dec [29]. The channel radius determines the device speed; lower radius corresponds to faster operation [26]. With respect to inversion mode devices, junctionless GAAFETs generally present higher I on /I o f f ratio and less short channel effects [25,30]. Regarding the low-frequency noise (LFN) behavior, junctionless GAAFET is almost not sensitive to gate bias and doping concentration variations [28]. Instead, the intrinsic gain and cutoff frequency were observed to be degraded by the hot carrier effect; a relative degradation of 15.44% for both of the analog parameters was reported [32]. The designer could improve the analog performance (small signal parameters and drain current drivability) by adding source and drain extensions, as shown in Figure 6 [24]. The structure that is depicted in Figure 5 can be further modified in order to increase the device performance. For example, a gate insulator made of hafnium oxide (H f O 2 ) instead of silicon oxide (SiO 2 ) can enhance the DIBL and the SS [27]. Apart from engineering the gate oxide structures, the designer could also increase the number of gates. If two gates are present, the structure is defined as a twin gate transistor (Figure 7). Such a structure allows implementing logic gates easily since two inputs are present [34]. The twin gate structure can also be applied to a double channel GAAFET, as shown in Figure 7b. A fabricated twin gate double channel GAAFET showed an I on /I o f f ratio of 7 × 10 8 , a DIBL of 83 mV/V, and a SS of 105 mV/dec [33]. Besides silicon and polysilicon channel junctionless GAAFETs, devices composed of other materials were also reported: a gallium arsenide junctionless GAAFET was simulated, leading a SS value near to the theoretical limit (58.2 mV/dec at 293.15 K) [27]. A germanium junctionless GAAFETs was compared to a silicon one, and it provided lower DIBL, SS and I on /I o f f ratio (data in Table 1) [35]. The channel material composition also influences the threshold voltage sensitivity to the temperature: considering silicon, gallium arsenide, indium arsenide, and indium phosphide, the minimum and maximum threshold voltage variations were observed for indium arsenide and silicon, respectively [37]. It is possible to use strain technology in order to further increase the device performance; a layer of SiN is deposited and, depending on the deposition conditions, the strain could be compressive or tensile [36]. The figures of merit of reported junctionless GAAFETs are presented in Table 1.  Figure 8a shows a schematic of a bulk junctionless FinFET. The device can also be fabricated on the top of an insulator layer, as shown in Figure 8b. In that case we define it as an SOI FinFET. Dimensions of these transistors strongly affect their performance. Considering bulk junctionless FinFETs, increasing the fin width (W) from 6 nm to 15 nm can lead to a variation of approximately 60% and 42% for DIBL and SS, respectively; changing the gate length (L G ) from 12 nm to 21 nm can lead to a variation of approximately 52% for DIBL and 14% for SS [12]. Variations in the fin height (H) are more critical in terms of analog performance [19]. With respect to the inversion mode device, the junctionless FinFET presents lower I o f f . This is attributed to the low carrier concentration and high electric field in the middle of the channel in the OFF state. As compared to SOI FinFETs, the bulk structure presents an additional degree of freedom in the design: by varying the doping concentration of the substrate from 1 × 10 18 cm −3 to 1 × 10 19 cm −3 , a change of 30% in the threshold voltage can be obtained [4]. Furthermore, it also provides lower SS and DIBL. From an analytical point of view, it can be modeled from conventional triple gate (TG) structures [3,20]. An alternative structure is the so-called SON (Silicon On Nothing). In this structure, the silicon layer is isolated from the substrate through a selective etching. This design choice implies a more complex fabrication process with respect to the bulk structure [5]. The designer could follow different approaches in order to optimize junctionless FinFETs: work function engineering of the gate to reduce I o f f (by changing the gate work function from 4.5 eV to 5.4 eV, I o f f can be reduced by five order of magnitudes) [7]; spacer engineering to improve performance (e.g., dual-k spacers architecture can provide an improvement in I on by 72.5% and in DIBL by 37.8%) [9]; doping engineering by using a Gaussian doped channel, which can lead to an increase in I on by 21.1% [10,13], or a lightly doped channel, which allows for better gate control on the device [11]; gate oxide engineering to provide higher performance (in terms of I on /I o f f and DIBL) by the implementation of complex hetero gate oxide structures [8]. For example, the double hetero gate oxide (DHGO) presented in Figure 9 can obtain a higher I on /I o f f with respect to conventional and triple/quadruple hetero gate oxide (THGO/QHGO) structures. Besides the number of hetero gate oxides regions, it is also important to consider their dielectric constant value: for high values (k = 40) the DIBL is reduced and the analog performance is degraded [14]. Instead, dual-k structures with intermediate values of the dielectric constant (k = 22, H f O 2 ) provide better performance with respect to those with low dielectric constant (k = 3.9, SiO 2 ) when considering random dopant fluctuation in the fin [16]. The latter is a critical phenomenon, especially for junctionless FinFETs, since they are more likely to be affected by random dopant variability [17]. Moreover, random dopant fluctuations and work function variability are considered to be more dominant with device dimension scaling [18]. It is important to note that the results reported for the junctionless FinFETs do not necessarily hold for the other junctionless transistors: for instance, the threshold voltage of junctionless FinFETs is more sensitive against work function variations as compared to that of junctionless GAAFETs [15]. FinFETs were analyzed by considering different materials for the device layer. A FinFET made of polycrystalline silicon is reported as a cost-effective solution with respect to silicon devices [22]. A GaAs FinFET with I o f f of 1 × 10 −15 A compared to a silicon FinFET with an I o f f of 1 × 10 −8 A was proposed. The better performance was attributed to the higher depletion of carriers when the device is in the OFF state [21]. More complex structures were reported: InGaAs junctionless FinFET with alloyed Ni-InGaAs source and drain [23]. Table 2 presents the main figure of merits of the reported FinFETs.

Single Gate
The single gate junctionless transistor presents two types of structures, i.e. bulk and SOI, as shown in Figure 10. The bulk structure provides more control on the device characteristics because of the possibility to dope and bias the bulk well [38]. When considering an n-type JLT with p-type bulk, it is possible to improve the hot carrier effect, thus reducing the I o f f current by positively biasing the well. On the other hand, by increasing the bulk bias, the threshold voltage can be decreased while increasing DIBL and SS. The degradation is even more relevant if the channel length is below 20 nm [40]. If the substrate doping concentration is high, then the I o f f current is minimized [44]. Moreover, it has to be considered that bulk junctionless transistors present reduced effective thickness, as compared to SOI; if a bulk SGJLT has a physical thickness of 10 nm, the effective thickness is 5 nm, because of the built-in junction potential [38]. With respect to the SOI structure, the bulk SGJLT presents improved analog performance: improved output transconductance, output resistance, Early voltage, and intrinsic gain [39]. When compared to junction transistors, the junctionless ones were observed to be more sensitive to the T Si /W Si ratio, and to provide a lower I on . This is attributed to the highly doped channel, which increases the scattering effect, thus lowering the mobility [42]. More options are available to the designers in order to optimize the SGJLT; gate work function engineering can lead to an improvement of 29% in the I on /I o f f ratio [41], while a non-uniform (Gaussian) doping concentration through the device layer can reduce the I o f f [43].
As for FinFETs and GAAFETs, an improvement of the electrostatic characteristics of the transistor can be obtained by implementing high-k spacers, as shown in Figure 11a. The high-k spacers enhance the fringing electric fields; as a result the device is depleted not only below the gate but also laterally. This implies an increment of the effective channel length, which, in terms, improves the SS [45]. Designers can also enhance the transistor mobility and currents through S/D engineering and dual-k spacers structures [46,47]. A variant of the conventional SOI structure is the SELBOX (selective buried oxide), as shown in Figure 11b. This oxide configuration improves the transistor thermal isolation, allowing for an increment for the I on /I o f f of 6 orders of magnitude (from 2.31 × 10 3 to 1.5 × 10 9 ) [48]. Moreover, the SELBOX architecture increases the gate control on the device, since a pn junction is formed between the highly doped channel and the substrate. The junction enhances the device layer depletion. Designers could also add a metal layer on the top of the BOX layer. This enables the formation of a Schottky junction that could help to fully deplete the transistor in the OFF state [50]. Regarding leakages in junctionless transistors, one of the most critical cause is associated to the parasitic BJT (bipolar junction transistor), as depicted in Figure 12. As electrons tunnel from the valence band to the conduction band (band-to-band-tunneling), they leave holes in the channel that can raise its potential. This phenomenon triggers a parasitic BJT between the source, the channel, and the drain in the OFF state [49].
The holes that accumulated in the floating body of the channel can cause a forward bias of the junction associated to the source/channel regions; if this bias turns on the parasitic BJT, then a large leakage current is observed. A possible solution is to employ thin film transistors (Section 3.6), which can reduce the band-to-band-tunneling and, therefore the associated leakage.  Figure 13 shows the structure of a double gate junctionless transistor. Many models were proposed, and the difference among them depends on the approximations that are involved in the derivation and the considered effects. For instance, many models do not consider short channel and quantum effects [53,55,58,66], while others are only valid for certain doping concentrations and device layer thickness ranges [59,61]. Quantum effects are critically important, because they can affect the threshold voltage [74]. The main issue is modeling the transition between the depletion and the accumulation regions, since the physical behavior is not the same in the two operating regions [56]. A technique involving high doping concentration in the device layer can be considered to reduce the model complexity. This assumption allows for simplifying the depletion width modeling or using the separation of variable in the Poisson equation [62,67]. A model that describes the current in all of the conduction regimes was proposed in [51]. It was validated for symmetrical long channel DG JLTs and describes the device behavior with a continuous current model. Regarding p-type devices, a threshold voltage model was proposed by [52]. They observed that the threshold voltage increases in absolute value as the device layer thickness, the doping concentration and the oxide thickness are increased. Extracting the threshold voltage and the current is therefore important in order to decide the doping concentration, and the gate oxide and device layer thicknesses [54]. Regarding the device performance, it could be negatively affected by the BTBT, which increases the leakage current. A design choice that improves the performance is the implementation of a thicker gate oxide near the gate edges ( Figure 14). It was observed that by modifying the gate oxide structure, the energy bands of the carriers under the gate are modified as well, resulting in a reduction of the leakage current [69]. The leakage current is lower in double gate junctionless transistors than in SGJLTs [43]. As reported in Table 3, the DGJLT presents better performance both for uniform and non-uniform (Gaussian) doping concentrations. Designers could also implement stacked-oxide structures. When compared to the conventional architecture, they present higher I on /I o f f , lower SS and DIBL [57,65]. By choosing a high dielectric constant material (e.g., H f O 2 ), a reduction of the leakage current as well as an improvement of the analog parameters could be observed [63,70]. Besides gate oxide engineering, spacer engineering could lead to a performance improvement. Spacers have an influence on the lateral extension of the depletion width and, therefore, on the effective channel length [72,73]. A simpler approach is doping concentration engineering. It was reported that a concentration of 1 × 10 18 cm −3 can significantly reduce the threshold voltage sensitivity by 70-90% with respect to the device layer and gate oxide thickness [68]. Graded doping profile can reduce I o f f by six orders of magnitude [71]. To correctly model JLTs, it is important to also model the carrier mobilities. The main issue is that the bulk mobility is lower than the accumulation one, because of screening effects. The accumulation mobility can be extracted by taking the second derivative of the 1/I acc curve. The bulk mobility can be computed by knowing the flat band voltage [64]. The mobility values can be degraded in case high voltages are applied [60]. Moreover, the implementation of complex equations (Schrödinger) is required, as well as the knowledge of parameters, such as impurities and surface roughness scattering mechanisms [75].

Thin Film
Thin film junctionless transistors are characterized by an ultra-thin channel thickness (≤10 nm) and very high doping concentration (≥1 × 10 19 cm −3 ). The thin film is better for obtaining the full depletion in the OFF state, while the high doping concentration ensures high current to flow in the device [77]. An important characteristics of these transistors is their device layer material composition; the majority of reported thin film transistors has polycrystalline silicon as channel material [79,82,84,85]. Therefore, they are identified based on the channel thickness (ultra-thin) and material composition (polysilicon). When considering polysilicon instead of silicon, an important difference arises: the polycrystalline silicon is composed of many crystallites connected by grain boundaries, as shown in Figure 15a. When electrons get trapped in these boundaries (Figure 15b), a space charge potential Φ B is formed. The stability of this potential depends on the applied drain voltage. If the latter is too high, trapped electrons could become unstable. The electrons instability influences the grain boundaries potential, which causes oscillation in the drain conductance. This phenomenon is more critical in TFTs with double gates, since the higher mobility allows for the electrons to easily destroy the trapped ones, and increase the oscillation [81]. Designers can limit this phenomenon by increasing the doping concentration [76].
When compared to junction TFTs, the junctionless ones present smaller transconductance g m and drain conductance g d . This implies larger Early voltage, improved low frequency noise and higher signal-to-noise ratio (SNR) [78,83]. The performance of thin film junctionless transistors mainly depends on the film thickness and the doping concentration. A high doping concentration ensures high I on , but it also lowers the SS since high carriers concentrations could screen the electric field induced by the gate. The SS also decreases with reduction in device layer thickness [86]. In addition, the temperature can significantly affect the device performance; when the temperature increases, the threshold voltage decreases (in absolute value) and the SS increases. This is attributed to the fact that the energy band gap E g decreases with temperature, thus increasing the carrier concentration [80]. Table 4 presents the figure of merits of the reported thin film junctionless transistors.  3.7. Tunnel FET Figure 16 shows the structure of a junctionless tunnel field effect transistor. The device is uniformly and highly doped. The middle gate acts as a control gate, while fixed voltages are applied at the side gates. When considering an n-type device, the tunneling effect can be triggered by correctly fixing the voltages on the side gates, where the source, channel, and drain regions (n-n-n) are converted into a (p-i-n) structure. When a certain control voltage is applied, the barrier between the source and the channel becomes narrower. As a result, current flows because of tunneling. Therefore, the conduction mechanism is different with respect to the other JLTs, since it is not based on depletion. The high-k dielectric below the gate (Si 3 N 4 , k = 7.5) improves the internal electric field, and, thus, the gate control [94]. The low-k spacers (SiO 2 , k = 3.9) are used to isolate the gates; by increasing the dielectric constant of the low-k spacers, it is possible to reduce I o f f [87]. Increasing the device layer doping concentration leads to an increment of both I on and I o f f , with the latter being more sensitive to doping variations. Decreasing the doping concentration leads to an improvement of the SS, since its value decreases from 290 mV/dec to 47 mV/dec as the doping concentration decreases from 2 × 10 19 cm −3 to 1 × 10 19 cm −3 . Therefore, one of the main advantages of junctionless TFETs is the possibility to achieve sub 60 mV/dec SS. Channel length reductions cause an increment of DIBL, and so of the I o f f [87]. Regarding double gate junctionless TFETs (Figure 17), the increment of the dielectric constant k leads to an improvement of I on . Increasing the insulation layer thickness causes an improvement in both I on and SS. However, this design choice leads to an increment of the parasitic capacitances [88,92]. To improve the robustness of junctionless TFETs, it is possible to selectively introduce dielectric materials in the gate oxide, which can reduce the variations in the coupling capacitance, allowing for higher immunity in terms of sensitivity [93]. The performance can be further increased by implementing dual-material gate (Figure 17a) or heterojunctionless structures (Figure 17b). The energy bandgap of these structures leads to higher I on and I on /I o f f , and lower SS [89][90][91]. Besides silicon, other materials were used for the device layer: a junctionless TFET made of indium arsenide was proposed [96]. The figure of merits of the reported junctionless transistors are collected in Table 5

Conclusions
In this work, junctionless transistors that were proposed over the last decade were studied. In particular, the influence of the technological parameters on the main figure of merits (I on /I o f f , DIBL, and SS) were analyzed. Design techniques, such as oxide/doping/spacers engineering, have been reported. Depending on the design choices, all of the typologies of junctionless transistors can present a high I on /I o f f ratio, as well as quasi-ideal subthreshold swing and optimal values of DIBL. Therefore, it is not possible to determine the best junctionless transistor solely based on the performance parameters. In general, a flexible optimization is associated to the more complex structures. The TFET is difficult to design, since the gate voltages need to be set carefully, and the work function differences must guarantee the tunnel behavior. The DGJLT has a less complex structure compared to TFET, and it presents an additional degree of freedom and enhanced gate control with respect to SGJLT and TFT. The SGJLT has a simple structure, but its performance is not comparable with the other junctionless transistors because of its reduced flexibility in terms of structure engineering. The TFT only presents high performance parameters if the device layer is highly doped and made of very thin polysilicon. FinFETs provide more flexibility in terms of structure engineering as compared to nanowires. The electrostatic control of both nanowires and FinFETs can be increased by surrounding the entire channel with the gate (GAA configuration). Regardless of the structure, junctionless transistors present easier fabrication process and competitive performance when compared to the junction transistors. As junctionless transistors are capable of reaching quasi-ideal subthreshold swing, optimal DIBL values, and high I on /I o f f ratio, it is expected that they will replace junction-based electronic devices in the following decade. Junctionless transistors are, therefore, the main candidates to become the conventional field effect transistors of the future.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

Abbreviations
The following abbreviations are used in this manuscript: