2.1. ESD Protection Structure and Mechanism
To overcome the fundamental challenge associated with conventional in-Si PN-based ESD protection structures, a novel above-IC graphene-based NEMS switch device structure was proposed and demonstrated experimentally [
11]. The new gNEMS ESD switch structure is illustrated in
Figure 4, which is a two-terminal device containing a suspended graphene nanoribbon over a cavity in a substrate. The two electrodes, i.e., the anode (A) and the cathode (K), are electrically separated by the cavity, hence in a normally OFF state. For on-chip ESD protection, a gNEMS device is connected to pads on an IC chip similar to that used in conventional in-Si PN-based ESD protection structures. An OFF gNEMS ESD switch will not affect normal IC operations. During an ESD event, when an ESD transient appears at one pad, the strong, transient electrical field generated will pull the suspended graphene film downward toward the bottom of the cavity. When the graphene film touches the cathode, it turns the gNEMS ON and forms a low-R conduction path to discharge the incident ESD pulse and hence provides ESD protection. After the ESD transient is over, the strong elastic force of the graphene film will pull itself upward and return to its suspension state, thus turning the gNEMS switch OFF [
11,
12]. Graphene materials are considered for ESD protection due to their material properties, such as extremely high carrier mobility (~15,000 cm
2/Vs), very high thermal conductivity (4.84~5.30 × 10
3 W/m), high Young’s modulus, superior mechanical strength, and superlight weight [
13,
14,
15], all of which are desirable for ESD protection functions. For example, high carrier mobility results in low ESD resistance. Good thermal conductivity reduces ESD-induced overheating. High Young’s modulus and light weight ensure the fast switching speed of gNEMS devices. Strong mechanical strength increases the reliability of gNEMS ESD switches.
Compared with conventional in-Si PN-based ESD protection structures, this new gNEMS ESD switch device has several novelties and advantages [
12]: First, the gNEMS device contains an air cavity and does not have any PN junction, which minimizes the ESD-induced parasitic capacitance and leakage current, ideally C
ESD ⇒ 0F and I
leak ⇒ 0A. This attribute is critically beneficial to advanced ICs because parasitic C
ESD can seriously affect RF ICs, and I
leak increases standby power consumption. Second, the ultra-high carrier mobility of graphene means the gNEMS device can carry more ESD current without overheating at a faster speed, translating into high ESD protection capability. Third, superior thermal conductivity facilitates the removal of the ESD-induced heat and hence enhances ESD robustness. Fourth, gNEMS is made in the back-end of the line (BEOL) in a complementary metal oxide semiconductor (CMOS), instead of residing inside a Si substrate, which theoretically removes the troublesome ESD-induced design overhead effects that are inherent to in-Si PN-based ESD protection structures. This novel above-IC gNEMS ESD switch structure can not only minimize the PN-induced ESD parasitic effects but also (ideally) consume no extra Si die area, which will also make IC layout floor planning much easier. Fifth, a gNEMS device is ideally a symmetric structure that can discharge ESD pulses in both directions, which can dramatically reduce the total head count of ESD protection devices on a chip required to form an ESD protection network. Sixth, a gNEMS device can be fabricated using CMOS-compatible processes. Overall, this disruptive gNEMS ESD protection device concept has the potential to revolutionize the ESD protection design field in the future.
2.2. gNEMS Fabrication
A fabrication procedure, depicted in
Figure 5, was developed for gNEMS devices considering both CMOS process compatibility and 3D heterogeneous integration for future ICs.
Figure 5a illustrates the five key processing steps for fabricating gNEMS devices. The substrate used is a heavily P-doped silicon wafer. First, low-pressure chemical vapor deposition (LPCVD) is used to deposit ~250 nm thick silicon dioxide (SiO
2) above the doped silicon wafer as the main dielectric layer. Second, a thin layer (100 nm) of silicon nitride (Si
3N
4) is deposited using the plasma-enhanced chemical vapor deposition (PECVD) method as a hard mask for hydrogen fluoride (HF) etching. After Si
3N
4 deposition, reactive ion etching (RIE) is used to open a trench window in the substrate. Next, a graphene film is grown on the copper via chemical vapor deposition (CVD) with the process optimized for fabricating large-area graphene films for production.
Figure 5b illustrates the Ramen spectra of both single-crystalline and poly-crystalline graphene films generated, in which the G and 2D peaks confirm the graphene structure, while the D peaks distinguish the structures of poly-crystalline and single-crystalline graphene materials. In the next step, the graphene film is transferred to the Si substrate, and oxygen plasma etching is used to pattern the graphene film into individual ribbons for gNEMS devices. Next, Ti/Pd/Au films (5/30/50 nm) are deposited using an e-beam, followed by a lift-off process to form the top electrodes. In the last step, the HF vapor method is used to etch off the SiO
2 within the Si
3N
4 widow and release the graphene ribbons.
Figure 5c shows a 3D scanning image for a fabricated gNEMS device structure, in which the suspended graphene ribbon can be readily observed [
11,
12].
2.3. Simulation Study of gNEMS Devices
To understand the gNEMS ESD protection mechanism and guide gNEMS design, a finite element method (FEM)-based simulation was conducted.
Figure 6 shows a gNEMS ESD device in an ESD-test setting using a transmission line pulse (TLP) ESD stress tester. The simulated gNEMS had a length of (L) = 20 µm, a width of (W) = 10 µm, and a cavity depth of d = 350 nm.
Figure 6a shows the vertical physical displacement (
Z-axis) of the graphene ribbon at the moment when the suspended graphene ribbon touched the bottom electrode. The vertical displacement of the graphene ribbon is scaled in colors, with blue for “0” displacement (i.e., the suspended graphene film in its original position) and red for the largest bending displacement at the center (−350 nm bending). The bending and contact of the graphene ribbon appear to be uniform across the ribbon width.
Figure 6b depicts the simulated vertical displacement characteristics of the graphene ribbon in the time domain during the TLP stressing period under a square pulse waveform of 7.2 V. It is readily observed that the suspended graphene membrane has the largest bending at the center, and the physical displacement increases as the TLP pulse continues in the time domain until touching the bottom. The simulation shows that as the time elapses, the electrostatic force induced by the TLP pulse will pull down the graphene ribbon at the central part. Once the graphene ribbon starts to bend, the intrinsic elastic force appears in the graphene membrane. As the distance between the center of the graphene ribbon and the bottom of the cavity decreases, the TLP-induced electrical field force increases to pull down further, while the elastic recovery force in the graphene will also increase. Since the ESD-induced electrostatic force is much stronger than the elastic force, the suspended graphene ribbon will continue to bend until it touches the bottom electrode to turn the gNEMS switch ON for ESD protection. This simulation helps to optimize the gNEMS design for which the ESD-induced pull-down force and elastic recovery force are considered to ensure the gNEMS ESD switching function. After the TLP pulse is over, the ESD-induced electrostatic force will immediately disappear, and the intrinsic elastic force will dominate and pull up the bent graphene ribbon back to its original position, thus turning the gNEMS OFF [
16].
FEM simulation can also be used to investigate the stress effect of the suspended graphene ribbon in a gNEMS under ESD zapping and hence provide design guidelines for improving the mechanical reliability of gNEMS devices. As shown in
Figure 4, the graphene ribbon was held by metal pads on both ends. During ESD actions, the ESD-induced electrostatic pull-down force creates stress on the graphene ribbon, particularly under the metal pads. In extreme cases, a physical fracture may occur, causing the mechanical failure of a gNEMS structure. To analyze the graphene fracture stress behaviors, various “nails” are designed to “hold” the graphene ribbon at the pad locations, as depicted in
Figure 7 [
17]. For a comparison study, four nail design splits were designed: a single square nail, a single triangular nail, and four square and triangular nails, which pin down the graphene ribbons with the pads. The idea for the nail design splits was that the nail shape affects the stress, and having more smaller nails may mitigate the stress effects.
Figure 8 depicts the fracture stress maps generated through FEM simulation, with the mechanical stress intensity color-coded as blue to indicate the lowest and red to indicate the highest stress pressure. It is readily observed that a single nail induces much heavier stress over the four-nail cases. The single square nail tends to have more stress than its triangular nail counterpart. On the other hand, the case with the four triangular nails is subjected to the lowest stress.
Table 1 summarizes the fracture stress results for the four nail designs.
2.4. Experiment Results for Poly-Crystalline gNEMS
The gNEMS prototypes were initially designed and fabricated using poly-crystalline graphene ribbons including varying design dimensions [
11,
18].
Figure 9 depicts the DC-measured I–V characteristics for sample gNEMS devices with varying graphene ribbon lengths of L = 7 μm, 10 μm, 15 μm, and 20 μm, respectively. A diode-like I–V curve shows the turn-on feature of the gNEMS devices. It is also observed that the turn-on voltage is dependent upon the graphene ribbon length, which is reasonable since a longer graphene ribbon undergoes a stronger pull-down force and a weaker elastic force and hence has more potential to bend, leading to a lower turn-on voltage. The current compliance was set to 0.1 mA in DC testing to avoid device failure.
A transient ESD stress test was then conducted using a TLP tester featuring a rise time of 10 ns and a pulse width of 100 ns.
Figure 10 depicts the measured ESD I–V characteristics for a sample gNEMS device (d = 350 nm, L = 7 μm, W = 5 μm) under both TLP stressing directions. The transient ESD behavior is clearly achieved. More importantly, the I–V characteristics of dual-directional ESD are observed for the gNEMS devices, which is a unique feature of this gNEMS ESD switch. The slight difference in the I–V curves in two opposite directions is attributed to the imperfection of the gNEMS prototype, as shown in
Figure 4. TLP testing confirms that the gNEMS stays OFF until the TLP pulse increases to a certain high level, which will quickly trigger the gNEMS into a low-R discharging mode for ESD protection. Experimental results show that the ESD-triggering voltage (V
t1) can be adjusted by device design parameters, including cavity depth, as well as the width, length, and shapes of graphene ribbons. The measured leakage current is very low, ~3–13 pA. This gNEMS can handle very high ESD current of up to ~10
8 A/cm
2, equivalent to ~1.5 KV/μm
2, which is much higher than ~7.5 V/μm
2 for an SCR ESD device (normally considered the most robust ESD protection devices in traditional PN-based structures). It is worth noting that gNEMS supports ESD in both directions, which can dramatically reduce the total ESD head count on a chip and thus significantly reduce the problem of ESD protection design overhead.
The temperature dependence of the ESD behaviors of gNEMS was investigated via TLP testing.
Figure 11 shows ESD I–V curves for a sample gNEMS device (L = 10 μm and W = 3 μm) under TLP stress at different temperatures, i.e., T = −10 °C, 30 °C, and 110 °C. It is readily observed that the gNEMS behavior is sensitive to temperature. As temperature increases, V
t1 decreases. Higher temperature also affects the current handling capability of gNEMS due to thermally induced defects in the graphene membrane [
16].
2.5. Experiment Results for Single-Crystalline gNEMS
Since the material properties of graphene films will affect gNEMS device performance, single-crystalline graphene was developed to improve the performance of gNEMS devices. A comparison study was carried out for poly-crystalline and single-crystalline graphene gNEMS devices.
Figure 12 depicts the I–V characteristics of ESD for the single-crystalline gNEMS devices under both TLP and very-fast TLP (VFTLP) stress tests.
Figure 12a presents the DC sweeping test result for a sample gNEMS (L = 5 μm and W = 3 μm) where a diode-like turn-on I–V curve is clearly observed with the turn-on voltage of ~2.45 V.
Figure 12b shows the I–V curve of ESD under TLP testing (t
r = 10 ns and t
d = 100 ns) for the evaluation of human body model (HBM) ESD. The transient ESD I–V curve is observed with V
t1~7.79 V and I
t2~30.3 mA. The leakage of I
leak ~2 pA is negligible.
Figure 12d depicts the ESD voltage and current behaviors in the time domain during TLP stress, based on which the ESD response time (t
1) of gNEMS can be obtained.
Figure 12c illustrates the I–V curve of ESD for gNEMS under ultra-fast VFTLP stress test (t
r = 100 ps and t
d = 1 ns), which clearly shows that the gNEMS can respond to the ESD pulses of the ultra-fast-charged device model (CDM). According to the VFTLP test results, V
t1~4.2 V and I
t2~31.3 mA are observed for the gNEMS device [
12,
16].
A comparison of the ESD characteristics of single-crystalline and poly-crystalline gNEMS devices is given in
Figure 13. In both DC sweeping and TLP stress tests, it is revealed that single-crystalline gNEMS outperforms its poly-crystalline counterpart, for example, with I
t2~0.37 mA for single-crystalline gNEMS over I
t2~0.14 mA for poly-crystalline gNEMS in the DC test. This is also reflected in I
t2~31.1 mA for single-crystalline gNEMS over I
t2~5.88 mA for poly-crystalline gNEMS in the TLP stress test [
12]. The performance enhancement of single-crystalline gNEMS is mainly attributed to the better material properties of single-crystalline graphene films as a result of which defects are dramatically reduced, and the crystalline grain improves both the electrical and thermal conductivity of graphene.
The reduced defect density in single-crystalline graphene can significantly improve the reliability of gNEMS devices, which was confirmed through repeated ESD stress tests, and their results are shown in
Figure 14.
Figure 14a depicts the results of the DC sweeping test repeated 11 times for a single-crystalline gNEMS where the DC sweeping voltage is clamped below the thermal breakdown current threshold (~0.24 mA) to avoid device failure. This repeated testing approach ensures that the same gNEMS sample can be used for repeating tests, thus increasing the reliability of the analysis results. It is readily observed that the I–V curves of the DC turn-on remain unchanged during the 11 times that the DC sweeping tests were repeated, indicating the good device reliability of the gNEMS switch.
Figure 14b shows that the single-crystalline gNEMS sample device has very stable ESD I–V characteristics after repeating the TLP stress test 110 times, which again confirms that the single-crystalline gNEMS device is very stable due to excellent crystalline graphene properties. During the repeated TLP stress tests, the TLP pulse was limited to under the thermal breakdown current (10 mA) to avoid device failure. For clarity, the I–V curves in
Figure 14b only show those after every 10 repeated stresses. Similarly,
Figure 14c shows ESD I–V characteristics for a sample single-crystalline gNEMS device subjected to the VFTLP stress test 110 times, which again clearly confirms that the gNEMS device is very stable owing to the good properties of single-crystalline graphene films [
12].
The impacts of the dimension of graphene ribbons on gNEMS performance were investigated using gNEMS devices designed with varying widths and lengths of graphene membranes. Both TLP and VFTLP measurements were conducted for a large number of gNEMS samples for statistical analysis.
Figure 15a depicts the statistical results of V
t1 ~ W for gNEMS devices of fixed L = 10 μm and varying W (3 μm, 5 μm, 10 μm, and 15 μm) under TLP stress. It is readily observed that V
t1 is not affected by the width variation because of the counter effect of an increase in both the electrostatic pulling force and the intrinsic elastic force as W increases.
Figure 15b shows V
t1 ~ L of different gNEMS devices with a fixed W = μm and varying L (5 μm, 7 μm, 10 μm, 15 μm, and 20 μm). The TLP test results clearly show that V
t1 monotonously decreases as L increases, because a longer graphene ribbon undergoes a stronger electrostatic pulling force while experiencing a weaker intrinsic elastic force, making it easier for the suspended graphene membrane to touch the bottom. Similarly,
Figure 15c,d shows that, under VFTLP stress, V
t1 is not affected by W but decreases for a longer L.
The impacts of graphene ribbon dimensions on the ESD current handling capability were also studied using a large number of gNEMS devices for statistical analysis, as depicted in
Figure 16.
Figure 16a shows the TLP-measured I
t2~W~L statistics for sample gNEMS, while
Figure 16b illustrates the same statistics for the VFTLP test. The test results show that I
t2 data range from 25.5 mA to 69 mA under TLP stress and from 27.6 mA to 59.9 mA for VFTLP results. It is readily observed that as W increases, at a fixed L, I
t2 substantially increases, implying that a wider graphene ribbon can handle more ESD current without overheating due to the reduced resistance. On the other hand, at a fixed W, I
t2 is generally not affected by L, and the slight decrease in I
t2 for longer L may be because a longer graphene ribbon may have more defects due to the imperfections generated during graphene growth.
Figure 16c,d show the highest I
t2 record measured in TLP and VFTLP tests for single-crystalline sample gNEMS devices (W/L = 7µm/20µm), i.e., I
t2 ~293 mA under TLP stress, or, the ESD current handling capability of J
t2~1.19 × 10
10 A/cm
2. Briefly, this is equivalent to an HBM ESD capability of ~178 KV/µm
2, which is much improved compared with that of poly-crystalline gNEMS, at J
t2~1.5 KV/µm
2; this is attributed to improved graphene properties in the single-crystalline structure. It is noteworthy that gNEMS is much more ESD-robust than any in-Si PN-based conventional ESD protection structures, e.g., J
t2~7.5 V/µm
2 for a typical SCR ESD protection device. A record high I
t2 of ~149 mA, i.e., J
t2~6.09 × 10
9 A/cm
2, was also obtained under VFTLP stress, indicating the superior CDM ESD protection capacity of gNEMS devices [
12].