Mechanism and Equivalence of Single Event Effects Induced by 14 MeV Neutrons in High-Speed QDR SRAM

Abstract

Single-bit upset (SBU) and multiple-cell upset (MCU) features of high-speed QDR-SRAM are revealed under the 14 MeV neutron irradiation. By comparing with the high-altitude real atmosphere test results directly, the equivalence of 14 MeV neutrons for atmospheric neutron-induced single event effect (SEE) evaluation is investigated. It is found that, compared with the 65 nm planar device, the SBU cross-section of 14 nm FinFET SRAM decreases to 1/58 and the proportion of MCU shows little difference, which results from the narrow channel between fin and substrate caused by shallow channel isolation in 14 nm FinFET process, and the charge sharing effect between fins is weakened. The SBU and MCU cross-sections under the 14 MeV neutron irradiation are underestimated by 22.8% and 85.7%, respectively. Besides, the probability and maximum size of MCU are both smaller than those in the real atmosphere. The MCU shape tends to be vertical, resulting from the smaller vertical spacing of sensitive volumes (about 100 nm). Further Monte-Carlo simulation shows that the total yield of secondary ions produced by atmospheric neutrons is higher than that produced by 14 MeV neutrons. Major Components of the “useful” products are p, Si, α, etc., which are the main cause of SBU events. Besides, compared with 14 MeV neutrons, atmospheric neutrons generate more kinds of secondary ions in the SV within the scope from p to W, and the diverse high-Z elements, such as W, Ta, Hf, etc., are the main cause of MCU events. Moreover, the maximum LET of secondary ions can reach 31.5 MeV·cm²/mg. The equivalence of using 14 MeV neutrons for atmospheric neutron-induced SEE evaluation is closely related to the critical charge of the device under test.

1. Introduction

Atmospheric neutrons are produced by the interaction between high-energy cosmic radiation (including galaxy cosmic rays and solar wind) and the earth’s neutral atmosphere [1], for example, the fluxes in atmospheric neutrons at the ground and 10 km flight height of Beijing city are 7.3 n/(cm²·hr) and 1068 n/(cm²·hr) [2], respectively. Atmospheric neutron-caused SEE in key ground electronics and avionics are becoming critical reliability issues because the smaller feature size of integrated circuits (IC) results in worse SEE characteristics [3,4], resulting in data error, functional interrupt, or even the crash of the electronic system.

Fourteen-nanometer FinFET devices have excellent performance, such as low power consumption, high integration and fast working speed; they have been gradually widely used in various electronic systems. Because the SEE-sensitive region and charge collection mechanism in FinFET devices are significantly different from the common planar devices, it is necessary to research the SEE in 14 nm FinFET devices.

The 14 MeV deuterium–tritium neutron generators have gained increasing interest in evaluating the neutron-induced SEE sensitivity of modern ICs, with the advantages of great availability, low cost, and high efficiency (a large number of test boards can be arranged around the neutron target, thanks to the 4π distribution of generated 14 MeV neutrons). However, due to the clear difference between the energy spectrum of atmospheric neutron (meV~GeV) and 14 MeV neutron, the equivalence of 14 MeV neutrons for the evaluation of atmospheric neutron-induced SEE has been a hot topic in the community [5,6,7,8,9]. In 2013, F. Miller et al. verified a 14 MeV neutron equivalence model based on the 90 nm SRAM process and found that the error range of the SEE cross-section induced by 14 MeV neutrons was less than two times [5]. In 2017, C. Weulersse et al. studied the scope of 14 MeV neutrons for the evaluation of atmospheric neutron-induced SEE. It was found that the single event upset (SEU) cross-section is in good agreement, while more multiple-cell upsets (MCU) and single event latches (SEL) were induced by high-energy neutrons [6]. At present, controversy still exists about the equivalence of 14 MeV neutrons in various types of devices. Moreover, published work used the test results at the spallation neutron source as the golden reference. The neutron energy spectrum, flux, and incident direction of spallation neutrons are inconsistent with the real atmosphere case, resulting in the deviation of the experimental result. In this paper, for the first time, high-altitude real atmosphere test results are used to verify the 14 MeV neutron test results, and the underlying mechanisms are investigated by further Monte-Carlo simulations.

In this work, the characteristics of SEUs and MCUs in high-speed QDR SRAM induced by 14 MeV neutrons are revealed and compared with a high-altitude real atmosphere test directly, based on our previous work on high-altitude tests [10], spallation source experiments [11] and α particle experiments [12] of the QDR SRAM. Internal mechanisms are researched in more depth via Monte-Carlo simulations.

2. Experimental Setup

The parameters of the device under test (DUT) have been exhibited in

Table 1. Parameters of the DUT.

DUT	Technology	Capacity	Core Voltage	Package
SRAM	65 nm MOS	144 Mb (8 Mb × 18)	1.8 V	BGA
SRAM	14 nm FinFET	128 Mb (8 Mb × 16)	0.8 V	BGA

. The maximum speed of the DUT can reach 550 MHz, with a capacity of 128 Mb. It is widely used in high-speed network communication, high-performance computing, imaging, and other fields. The DUT was not de-capped before the irradiation since that 14 MeV neutron can penetrate the package.

Figure 1 and Figure 2 show the irradiation experiment setup. The SEE testing system is designed using the motherboard—high-speed connecting line—daughterboard framework, to reduce the probability of a radiation induced abnormal motherboard. Prior to the irradiation, the checkerboard pattern had been inputted into the DUT. During the period of neutron bombardment, the contents in the DUT had been compared with the golden data at regular intervals. Detailed information on SEU, such as the error time, address and data, could be obtained in the irradiation experiment. Furthermore, the experimental temperature was maintained at 25 ± 5 °C, the SEL phenomenon did not appear and the currents were monitored during the whole test.

The 14 MeV neutron irradiation test was carried out at the Key Laboratory of Nuclear Data, Chinese Institute of Atomic Energy. During the experiment, 14 MeV fast neutrons were generated by the T(d, n)⁴ He reaction. The deuterium beam was accelerated by the high voltage multiplier and bombarded the tritium target. The neutron yield was about 3 × 10¹⁰ n/s, and the flux range at the sample position was 10⁴~10⁵ n/(cm²·s).

3. Results and Analysis

The main contents of this section are: (1) revealing the 14 MeV neutron-induced SEE characteristics in 65 nm high-speed and high-capacity QDR SRAM, focusing on the MCU probability, shape, and inner mechanism; (2) comparing the experimental results of 14 MeV neutrons, the high-altitude test and spallation source test.

3.1. SEU Cross-Section

3.1.1. Results and Comparison

Under the 14 MeV neutron irradiation, the effect types are mainly single bit upset (SBU) and MCU. It should be noted that one MCU is marked as one error in the calculation of the MCU cross-section. Compared with the 65 nm plane devices, the SBU cross-section of the 14 nm FinFET device is reduced to 1/58. There is no significant difference in the proportion of MCU between the two devices. For the 65 nm plane devices, the SBU accounts for 93.8%, and the MCU ratio is 6.2%. For the 14 nm FinFET device, the MCU ratio is 5.3% and the largest MCU involves 3 bits. In addition, several motherboard malfunctions occurred, resulting from the scattering of neutrons, which can be recovered after resetting.

The height of the FinFET device is 45 nm, the width of the FinFET device is 15 nm, and the distance between the two FinFET devices is about 35 nm. The Shallow Trench Isolation (STI) physically separates the FinFET devices. Compared with the 65 nm plane devices (the area of a single memory cell is 1 μm × 0.5 μm), the area of a single memory cell of 14 nm FinFET devices is reduced to only 0.37 μm × 0.18 μm, and the size of SEU sensitive region decreases accordingly, which is the main reason for the decreasing of the SEU cross-section of the 14 nm FinFET device under 14 MeV neutron irradiation. For the MCU, although the spacing of 14 nm FinFET devices is only about 35 nm, and the critical charge is reduced to sub-Fc, the STI narrows the channel between 14 nm FinFET and substrate, and the charge sharing effect between FinFET would be weakened, which is the main reason why the MCU proportion of the 14 nm FinFET devices does not change significantly.

Figure 3 shows the SEU cross-sections of the QDR SRAM of 14 MeV neutron, high-altitude, and spallation source tests. The spallation source test results were obtained at the temporary terminal (BL09) of the China Spallation Neutron Source (CSNS) [11]. Moreover, the high-altitude test results are demonstrated in the next section in detail. It should be noted that the MCU cross-section is obtained by counting one MCU as one event. It can be seen in Figure 3 that SBU and MCU cross-sections induced by 14 MeV neutrons are about 22.8% and 85.7% lower than those in the real atmosphere, respectively.

Figure 4 shows the neutron energy spectrums and distribution ratios of the high-altitude and spallation source tests. It can be seen that, in the high-altitude test, the proportion of neutrons with energy greater than 14 MeV is 58.2%. Due to the lower proportion (5.5%) of E > 14 MeV neutrons at the temporary terminal of CSNS, SBU and MCU cross-sections of the spallation source test in Figure 3 are both lower than those of the high-altitude test.

3.1.2. Description of High-Altitude Test

A high-altitude real atmosphere test was completed at the Yangbajing International Cosmic Ray Observatory, Lasa, China, at an altitude of 4300 m [10]. There were four test boards, which include 72 QDR SRAMs, were placed on a water-level platform. Over the course of the 153-day experiment, 39 SEUs were observed, of which SBU accounted for 76.9% and MCU accounted for 23.1%.

The above SEUs were induced by high-energy neutrons (E > 1 MeV), thermal neutrons, and α particles from the device package together. To obtain the SEU cross-section of high energy neutron, the thermal neutron and α particle sensitivity of the DUT must be investigated. First, we found that thermal neutrons have no influence on the DUT [11]. Second, α particle induced soft error rate, SER_α, of the DUT was 303 FIT/Mb. All the SEUs were SBU events [12]. Therefore, α particle induced SEU number during the high-altitude test, N_SEU,α, can be calculated:

N_SEU,α = SER_α × T × N_bit/(1024 × 1024 × 10⁹)

(1)

where T and N_bit are the effective test time (unit: hr) and capacity (unit: bit), respectively.

Furthermore, the SEU cross-section of high energy neutron during the high-altitude test, σ_{high altitude}, can be calculated:

σ_{high altitude} = (N_SEU − N_SEU,α)/(flux × T × N_bit)

(2)

where N_SEU denotes the total SEU number during the high-altitude test, flux means the high energy neutron flux at the test site (unit: n/(cm²·hr)).

3.2. MCU

Figure 5 shows the MCU ratios of 14 MeV neutron, high-altitude and spallation source test. Clear differences can be seen. Compared with the high-altitude test, the ratio of 14 MeV neutron-induced double cell upset (denoted by MCU2) is about two times lower. Under the 14 MeV neutron irradiation, the largest MCU involves only 3 bits. While in the high-altitude test, the largest MCU involves 9 bits. The overall proportion of MCU induced by 14 MeV neutrons is 6.2%, which is lower than that of the high-altitude test (23.1%). It can be seen that the MCU probability and maximum size are both underestimated under the 14 MeV neutron irradiation, which may affect the design and verification of the error correction code (ECC) hardening scheme.

The MCU shapes produced by 14 MeV neutrons are further extracted, based on reported MCU addresses and bitmap of the DUT. In

Table 2. MCU shapes induced by 14 MeV neutrons.

SEU Type	Shape	Count
SBU		436
MCU2		27
MCU3		2

, it can be seen that the MCU shape tends to be vertical. All the MCU2 exhibit vertical orientation. Figure 6 shows the layout of sensitive volumes (SV) of the DUT with a checkboard pattern. The drain of the OFF-state NMOS is regarded as the memory cells’ SV. It can be seen in Figure 6 that the SVs are gathering together. The lateral spacing between SVs (about 300 nm) is wider than the vertical spacing (about 100 nm), which is consistent with the above MCU shape feature.

4. Monte-Carlo Simulation

The aim of the Monte-Carlo simulation of neutron transport is to research the inner mechanism of previous experimental results. The secondary ions characteristics in the device SV, including the ion species, LET, and deposited charge, under 14 MeV neutron and atmospheric neutron irradiation are obtained and compared.

4.1. DUT Simulation Model

The 65 nm QDR SRAM simulation model (see Figure 7) has been built on the basis of the reverse-technique results of the DUT [10] and the experimental setup. The model includes six metallization layers and copper is the main metal material. Tungsten plugs are located between M0 and the active silicon layer.

Table 3. Memory cell size and SV parameters of the QDR SRAM.

DUT	Memory Cell	SV (x × y)	Depth of SV	LETth	Critical Charge
QDR SRAM	1 μm × 0.5 μm	0.2 μm × 0.19 μm	0.45 μm	0.22 MeV·cm2/mg [14]	1 fC [15]

shows the memory cell size and SV parameter of the 65 nm QDR SRAM. The drain of the OFF-state NMOS is regarded as the memory cells’ SV. The depth of the SV is set as 0.45 μm in consideration of the funnel length and diffusion process. Importantly, note that the DUT is very sensitive to the proton direct-ionization effect due to the low LET threshold, which is smaller than 0.5 MeV·cm²/mg.

4.2. Neutron Transport Simulation

In this simulation, the Geant4 toolkit is adopted, since it is able to simulate the neutron transport process [16]. Several publications studied the difference between Geant4 and MCNP and found that concerning the total neutron yield, the two toolkits agreed within 20%. [17,18]. To raise the simulation efficiency, the x × y scale of the model and the SV has the same scales. Under normal circumstances, about 1 × 10⁹ neutrons would strike the surface of the device in the model. The neutron energy levels in the simulation include 14 MeV and the atmospheric neutron spectrum in Figure 4 (E > 1 MeV). It should be noted that for the neutron source used in this work, almost all the neutrons are within ±1 MeV of the peak energy of 14 MeV. This energy spread was considered in the simulation and little impact on the result was observed. Thus, monoenergetic neutrons of 14 MeV are used in the following comparison for simplicity. For the SV of the devices, the properties of secondary ions, which include the ion species, the LET, and the density of deposited charges, have been obtained by measuring and simulating calculations. This information can help in determining the characteristics of the 65 nm planar device and the 14 nm FinFET SRAM under 14 MeV neutrons and atmospheric neutrons, respectively, as well as the effect of new feature additions.

4.3. Simulation Results

Figure 8 shows the 14 MeV neutron and atmospheric neutron-induced secondary ion species in the device SV. An obvious difference can be seen. The total yield of secondary ions generated by atmospheric neutrons is higher than that generated by 14 MeV neutrons. Additionally, it is important to note the logarithmic scale on the Y-axis in Figure 8; it is very clear that the main reaction products are n, p, Si, etc., of which protons are perfectly capable of leading to some soft errors in the QDR SRAM with a critical charge of 1fC (shown in Table 3). This reason may account for this phenomenon in Figure 3, which can be described as the 14 MeV neutron-induced SEU cross-section being lower.

Besides, compared with the 14 MeV neutrons, atmospheric neutrons are able to generate more kinds of secondary ions in the SV within the scope from p to W. Due to this wide energy spectrum, more high-Z secondary ions, such as W, Ta, Hf, etc., are produced by atmospheric neutrons. This reason may account for the phenomenon in Figure 3 and Figure 5 where the MCU cross-section, ratio, and maximum size induced by atmospheric neutrons are higher than those induced by 14 MeV neutrons, since heavier secondary ions with higher LET in Figure 8 are the main cause of MCU events.

Figure 9 further shows the LET features of the secondary ions caused by 14 MeV and atmospheric neutrons in the SV of devices. First, under atmospheric neutron irradiation, the maximum LET of secondary ions in the device SV can be achieved at 31.5 MeV·cm²/mg. While the maximum LET decreases to 7.8 MeV·cm²/mg under the 14 MeV neutron irradiation. Second, as the LET increases, the overall ratio decreases rapidly under the 14 MeV neutron irradiation. These results suggest that the usage of 14 MeV neutrons may underestimate or even miss SEE types with high LET_th, such as SEU in hardened devices.

Figure 10 displays the characteristic of deposited charges, Q_d, in the SV induced via 14 MeV and atmospheric neutrons. In the area of Q_d < 17 fC, the deposited charge probability under the 14 MeV neutron irradiation is slightly higher than that under the atmospheric neutron irradiation. While in the area of Q_d >17 fC, the apparent result of this is that atmospheric neutrons are capable of generating secondary ions with larger LET values, longer ranges, and thus more deposited charges in the device SV, with also higher probability. The secondary ion makes the maximum of the charge deposited reach 118.5 fC.

Therefore, we can conclude that the equivalence of 14 MeV neutrons for atmospheric neutron-induced SEE evaluation is closely related to the critical charge of the DUT. For the SBU response of the QDR SRAM in this work, with a critical charge of 1fC, the SBU cross-section induced by 14 MeV neutrons is approximately 22.8% lower than the value in the real atmosphere. It can be predicted that the difference will decrease as the feature size shrinks. However, for MCU and other SEE types with high critical charge, the discrepancy between 14 MeV neutrons and atmospheric neutrons will increase.

5. Conclusions

In this work, the characteristics and inner mechanisms of 14 MeV neutron-induced SBUs and MCUs in 65 nm high-speed and high-capacity QDR SRAM are studied and compared with high-altitude real atmosphere test results. It is found that, for the 14 MeV neutron irradiation, the effect types are mainly SBU and MCU. SBU accounts for 93.8%, the MCU ratio is 6.2%, and the largest MCU involves 3 bits. While for the high-altitude test, the largest MCU involves 9 bits, with SBU and MCU cross-sections 22.8% and 85.7% higher than those under the 14 MeV neutron irradiation, respectively. Compared with the 65 nm plane process, the SBU cross-section of the 14 nm FinFET SRAM decreases to 1/58, and the difference between the MCU proportions in the two kinds of devices is small. The reason is that the shallow channel isolation in the 14 nm FinFET process makes the channel between FinFET and the substrate very narrow, and the charge sharing effect between FinFET would be weakened. In addition, the MCU shape tends to be vertical, resulting from the smaller vertical spacing of SVs (about 100 nm).

Monte-Carlo simulations have been used in the experiment to investigate the characteristics of secondary ions in the device SV further, including ion species, LET, and deposited charge and could compare two experimental environments, 14 MeV neutrons and atmospheric neutrons. The total yield of secondary ions generated by atmospheric neutrons is higher than that generated by 14 MeV neutrons. The majority of the “useful” reaction products are p, Si, α, etc., which are the main cause of SBU events. Besides, compared with 14 MeV neutrons, atmospheric neutrons generate more kinds of secondary ions in the SV within the scope of p to W; the heavier products are the main cause of MCU events. Atmospheric neutrons with a wide energy spectrum are able to generate secondary ions with larger LET, longer range, and thus more deposited charge in the SV of the 65 nm QDR SRAM, with also higher probability.

The equivalence of 14 MeV neutrons for atmospheric neutron-induced SEE evaluation is closely related to the critical charge of the DUT. Additionally, for MCU and other SEE types with high critical charge, the discrepancy between 14 MeV neutrons and atmospheric neutrons will increase.