> Submission to IEEE-TNS <

1

Abstract—Non-volatile memory (NVM) technology highly

resistant to ionizing dose and radiation effects in general continue to be a challenge for space missions. Novel NVM nano-ionic technologiesknown as Conductive Bridging Random Access Memory (CBRAM) resistive circuits exhibit great promise for both high density memory and also exhibit high total ionizing dose resilience. In this work, it is discovered that CBRAM can be sensitive to high TID levels. However this novel technology can be radiation-hardened by process, which is demonstrated in this paper.

Index Terms— chalcogenide glass, total ionizing dose, memristors, programmable metallization cell, PMC, CBRAM, conductive bridging RAM, nanoionic memory, radiation effects, photo-diffusion, resistive switching, ReRAM, cation, radiation hardening by process, radiation hardening, non-volatile-memory, NVM.

I. INTRODUCTION

se of off-the-shelf non-volatile memory technology in space systems/applications remains a challenge.

Nowadays NVM applications are commonly implemented with floating-gate transistors (i.e. Flash technology) a technology sensitive to ionizing dose (TID) as well as single event effects (SEE). TID in the range of 100krad induce failures in both NAND and NOR Flash commercial devices [1]. Future space exploration missions and for example missions toward the moons of Jupiter will be extremely challenging from the point of view of radiation effects. Jupiter’s radiation belts are the most important in the solar system and TID of several hundreds of kilo-rad and even Manuscript received November 06, 2015. This work was funded by the Defense Threat Reduction Agency under grant no HDTRA1-11-1-0055. The authors would like to thank Dr. Jacob Calkins of DTRA for his support of this work.

Y. Gonzalez-Velo is with the School of Electrical, Computer and Energy Engineering, Arizona State University, Tempe, AZ 85287-5706, USA (e-mail: yago.gonzalezvelo@asu.edu).

A. Mahmud, W. Chen, J. Taggart, H. J. Barnaby, K. E. Holbert, are with the School of Electrical, Computer and Energy Engineering, Arizona State University, Tempe, AZ 85287-5706, USA.

M. N. Kozicki is with the School of Electrical, Computer and Energy Engineering and with the Center for Applied Nano-Ionics, Arizona State University, Tempe, AZ 85287-5706, USA.

M. Ailavajhala, M. Mitkova, are with the Department of Electrical and Computer Engineering, Boise State University, Boise, ID 83725, USA.

mega-rad are expected over the mission life. Therefore, those missions will require highly TID tolerant devices. Novel radiation-hardened NVM solutions exist, however they are typically low density memories using a radiation-hardened CMOS process with a novel type of memory cell such as MRAM cells [2], [3].

Resistance switching in materials and resistive memory devices (ReRAM, CBRAM) that derive from it have appeared recently as alternatives to flash-based solutions for highly integrated NVM as well as ultra-low power memory [4-7]. A first commercial standalone memory of the aforementioned resistive switching technology has been presented with CBRAM, a cation based resistance switching technology [4], [8]. Additionally, recent results have demonstrated the high tolerance of CBRAM memory cells to TID levels much higher than the levels withstood by floating gate transistors [9-11]. This allows envisioning CBRAM technology as a viable solution for high-density TID tolerant non-volatile memories to be used in future space missions.

II. II BACKGROUND

CBRAM cells are resistance switching elements using active metals such as copper (Cu) or silver (Ag) diffusing through a solid-state electrolyte [8], [11-13] to switch resistance from a high resistance state (HRS) to a low resistance state (LRS). Such devices are utilized in novel NVM circuits and have been demonstrated to have characteristics that could compete with flash technology [4-6], in part thank to their scaling capability, their very low-power operation and their compatibility with standard CMOS process (i.e. easiness of foundry implementation). Moreover, it is interesting to note that CBRAM cells can be used in a variety of other applications rather than just memory, among which novel electronic circuits that perform neuromorphic computation. CBRAM cells have been demonstrated to be extremely resilient to TID up to Megarad levels of absorbed dose [9-11]. This TID resilience has been demonstrated on both Cu-SiO2 devices [11] and Ag-GeS (both CMOS process compatible) [10], as well as Ag-GeSe devices [9]. It has also been shown [1] that the retention of 128 kbit CBRAM memory circuits is greater than traditional flash-based memory circuits, with no variation of supply current for TID up to 450krad [1. This is shown on the results presented in figure 1 from [1], where the standby current and the numbers of errors

Radiation Hardening by Process of CBRAM Resistance Switching Cells

Y. Gonzalez-Velo, Member, IEEE, A. Mahmud, Student Member, IEEE, W. Chen, Student Member, IEEE, J. Taggart, Student Member, IEEE, H. J. Barnaby, Senior Member, IEEE, M. N. Kozicki,

Member, IEEE, M. Ailavajhala, Member, IEEE, K. E. Holbert, Senior Member, IEEE, M. Mitkova, Member, IEEE,

U

> Submission to IEEE-TNS <

2

on the memory array of a commercially available CBRAM circuit are evaluated as a function of dose,. These characteristics are compared to the same characteristics measured on a Flash-based EEPROM device. It was observed that for CBRAM circuits no TID induced effect such as errors or variation of standby current occur, while the characteristic of the Flash-based device exhibit important changes.

(a) (b)

Fig. 1. a) Evolution of the standby current Istd-by with TID for CBRAM based circuits (open square symbol) and Flash-based circuits (open triangle symbol) from [1]. b) Normalized number of errors as a function of TID for CBRAM based circuits (open square symbol) and Flash based circuits (open triangle symbol) from [1].

This result was the first result obtained on commercially available CBRAM memory circuit, demonstrating the radiation tolerance of this technology, and validating results obtained previously on CBRAM cells [9-11].

A Comparison between TID induced errors observed on several NVM technologies has also been presented in [1], showing that CBRAM memory circuits withstand high level of absorbed dose using a commercial CMOS process (non rad-hard CMOS process). This is a difference with solutions provided by Cobham (former Aeroflex) and Honeywell MRAM solutions (see figure 2) [2-3].

Fig. 2. Normalized errors as a function of TID observed on different type of NVM memory technologies, from [1].

From the TID resistance observed on CBRAM is in the use of CBRAM cells in very high total dose missions might be a solution. This led to the study/investigation/interest on the behavior of CBRAM cells at very high TID.

In this work, very high TID effects on CBRAM memory cells are investigated. The first observation of TID-induced failure of a CBRAM cell is presented. Two different manufacturing processes have been investigated, and it is demonstrated that CBRAM radiation hardness by process can

be implemented by selecting appropriate processing parameters. Failure to effectively switch resistance’s cell is observed when an agglomeration effect occurs at the surface of the cells, revealing that radiation-induced electron-hole generation can activate the formation of an unintended phase/material that could be the cause of the observed failure.

In the first section, CBRAM cells investigated in this work are described. In the following experimental section, the resistance switching characteristics of CBRAM devices made with the two different processes are presented after exposure to gamma rays. In the last discussion section, the agglomeration effect observed after gamma ray exposure of the CBRAM cell is examined and a discussion is provided concerning its impact on the reliability of devices depending on their manufacturing process.

III. CBRAM CELLS: MATERIALS, STRUCTURE, DESCRIPTION

The CBRAM cells studied in this work are presented in Fig. 1, which shows a cross section (Fig. 3.a) and a top view microphotograph (Fig. 3.b) of an individual cell. The CBRAM cells investigated in this paper are “active metal/electrolyte/inert metal” devices made of a silver doped Ge30Se70 chalcogenide glass that acts as a solid-state electrolyte (Ag-ChG in Fig. 3.a). The cells have an active top silver electrode, and an inert bottom nickel counter electrode. In this work two different processes (labeled as processes A and B) have been used to manufacture two sets of devices. Both processes utilize an initial 60 nm thin layer of Ge30Se70 chalcogenide glass film thermally evaporated. Silver is deposited on top of this Ge30Se70 layer in order to obtain the solid-state electrolyte and the top electrode [12-14]. The ChG and silver are thermally evaporated in a Cressington 308 thermal evaporator according to description provided in [9].

(a)

> Submission to IEEE-TNS <

3

(b)

Fig.3: (a) cross sectional view of CBRAM cells implemented on top of an SiO2/Si wafer. A nickel blanket film forms the common bottom electrode to the devices that are implemented and patterned on top of vias wet-etched through an additional SiO2 layer. b) top view of a CBRAM cell

Different sets of CBRAM cells from both processes have been irradiated with cobalt-60 gamma-rays in the Gammacell 220 irradiator at Arizona State University. Exposures were conducted at room temperature and with sample electrodes left floating during exposure (no bias applied during exposure) [9]. A TID of 11.5 Mrad(Ge30Se70) has been reached on parts from process B and a maximum TID of 10 Mrad(Ge30Se70) was attained on samples from process A. Samples were step stressed, meaning that electrical characterizations were performed at several dose levels below the maximum dose in order to retrieve the evolution of the switching characteristics with increasing dose. The devices were electrically characterized on a probe-station by performing DC current-voltage sweeps with an Agilent 4156C parameter analyzer. The voltage between the electrodes is swept from –1 V to 1 V. The varying voltage is applied on the active anode of the CBRAM cell, while the inert cathode is fixed at 0 V. Current-voltage (I-V) measurements were performed prior to and shortly after the irradiations.

IV. EXPERIMENTAL RESULTS

Current-Voltage characteristics obtained on devices from process A (Fig. 4.a) and process B (Fig. 4.b) are different because the choice of fabrication steps yields solid-electrolytes (Ag-ChG layer in Fig. 3.a) with different conduction properties. It is observed in Fig. 4 that prior to radiation exposure, devices fabricated in either process demonstrate excellent resistance switching (see dashed line representing resistance values on Fig. 4), but the range of resistance that can be obtained for these two processes are different. With Process A, the devices can be switched at much lower currents and the resistance range is much larger. The erase process for process A parts is also faster. Lastly, the voltage at which the device is switched from its low resistance state (LRS) to its high resistance state (HRS) (i.e. the threshold voltage) is also slightly different for both processes.

Fig. 4: Current-voltage (red solid line, left y-axis) and resistance-voltage characteristic (blue, dashed line, right y-axis) (a) Process A, (b) Process B

A. TID testing of PMCs manufactured with two different processes

TID testing was conducted on samples from both process A and B. High resistance and low resistance states were extracted from the DC current-voltage characteristics after exposure. Devices from process A can be effectively switched after exposure to a TID of 10 Mrad. The HRS and LRS levels obtained from continuous cycling of one, representative, irradiated 10 µm diameter CBRAM cell from process A are presented in Fig. 5.

> Submission to IEEE-TNS <

4

Fig. 5: HRS (black open squares) and LRS (red open circles) measured 500 times on process A cell exposed to 10 Mrad.

The irradiated part was successfully switched 500 times before the test was stopped. Similar behavior has been observed on several parts from process A. Devices from process B were also tested after exposure to TID, and the results of a step stressed exposure are presented in Fig. 6. It is observed that the devices from process B can be switched for TID levels up to 5 Mrad(Ge30Se70), but once a TID of 6 Mrad(Ge30Se70) is reached, resistance switching failure occurs. This can be seen in Fig. 4 where data obtained on exposed devices (Fig. 6.a) are compared to data collected on control devices (Fig. 6.b). In Fig. 4.a, it is shown that at TID lower than 6 Mrad(Ge30Se70), the HRS and LRS values are well separated and the parts are effectively exhibiting resistance switching For TID values higher than 6 Mrad(Ge30Se70), both HRS and LRS are equal and higher than their original value, at very high resistance levels (see Fig. 4.a). When both HRS and LRS are equal, the device is not switching anymore. The relatively uniform response of the control parts demonstrates that irradiation is the cause of part failure in process B parts. These results suggest that process A is much better suited for operating in high ionizing dose environments than parts from process B.

Fig. 6: (a) ON and OFF resistance as a function of dose (bottom abscissa axis) and time (top abscissa axis) of a 10 µm diameter CBRAM cell. ROFF is represented with black open squares and circles and RON represented with red triangles and diamonds. (b) ON and OFF resistance as a function of dose of two unexposed control 10 µm CBRAM cells.

An examination of the Ag top contact surface after radiation exposure provides clues to the mechanism of failure for process B parts. Several crystallites or agglomerated areas appear on the top of process B cells as illustrated in Fig. 7.

(a) (b)

(c) (d)

(e) (f)

0 100 200 300 400 5001k

10k

100k

1M

10M

R

esis

tan

ce (O

hm

)

#Operation - #Cycle

HRS - Roff LRS - Ron

> Submission to IEEE-TNS <

5

(g)

Fig. 7: Microphotograph of process B devices after gamma rays exposure: (a) 500µm square side device, (b) 10µm circular via device. SEM views of process A after gamma rays exposure: (c) 500µm square side device, (d) 10µm circular via device.

This effect did not occur on the surface of process A parts. In Fig.5, microphotograph and scanning electron microscope (SEM) images are presented, showing these agglomerated features on top of the CBRAM cells. It can be observed on Fig. 7.a and 7.b, that these features appear on both small area and large area devices from process B (figure 7.a, 5.b, 7.e and 7.f), while not appearing on devices from process A (figure 7.c and 7.d). An SEM close view of these agglomerated features is provided in figure 7.g

In figure 8 SEM images and the corresponding electron dispersive spectroscopy (EDS) compositional mappings are provided. It can be observed on the top SEM images provided in Figs. 8.a and 8.c that the crystallites appear within the via or on the edges of the via. From Figs. 8.b and 8.d, the relative amount and location of Ag atoms is measured and extracted. It is found that these crystallites that appear after exposure are silver rich features.

(a) (b)

(c) (d)

Fig. 8: Top SEM views of process B devices (a), (c); top EDS mapping of Ag atoms obtained process B parts (b), (d). The EDS mapping reveal the presence of the Ag element, and the scale provided on the bottom of the figure shows the quantification of silver.

(a) top SEM view of a 10 µm diameter process B CBRAM cell exposed to 11 Mrad. (b) EDS mapping of the device from (a). More silver is present within the 10 um via.

(c) 500 µm side square CBRAM cell, outlined with the dashed light line. It is observed that several crystallites form at the edges of the via. (d) EDS mapping revealing that amount of silver present at the edges of the via.

V. DISCUSSION

In this work it was discovered that resistance switching of CBRAM cells can be impacted by ionizing radiation. This is the case for process B, particularly for TID higher than 4.5 Mrad(Ge30Se70). For devices fabricated in process B, a large increase in the resistance of the CBRAM cells results and this response seems to correspond to agglomeration in the device surface, leading to the build-up / formation of new phases in the anode contact. Dissolution of the surface contact is therefore thought to result in the failure of process B devices. For process A however, no agglomeration effects are observed and the cells still operate after a very high TID level of 10 Mrad(Ge30Se70) is reached.

The changes observed in the top Ag surface reveal a change in the silver layer roughness and a change in the silver atomic distribution/location as revealed by the EDS mappings. The formation of a new phase, a relocation of silver atoms might be leading to a radiation induced modification of the solid-state electrolyte and its electrolytic properties. Moreover, a loss of the contact, or increase of the contact resistance between the aluminum contact and the silver top layer, due to enhanced roughness could lead to devices behaving like open circuits. Results of further material characterizations will be reported in the final paper, to gain insights on the composition of the small crystallites, agglomerated zones.

However, despite the loss functionality of process B parts, it is worth noting that these CBRAM cells exhibit a very good radiation hardness compared to common non-volatile memory cell technologies. Moreover, in this work it is demonstrated for the first time that that radiation hardness by process is possible to enable much higher TID tolerance of CBRAM technology.

REFERENCES [1] Y. Gonzalez-Velo, H. J. Barnaby, M. N. Kozicki, C. Gopalan, K.

Holbert, “Total Ionizing Dose Retention Capability of Conductive Bridging Random Access Memory,” IEEE Elec. Dev. Let. vol. 35, n. 2, Feb. 2014

[2] Cobham Hi-Rel Memories (2015, Oct. 8) [Online]. Available: http://ams.aeroflex.com/pagesproduct/prods-hirel-mems.cfm

[3] Honeywell Microelectronics Nonvolatile MRAM (2015, Oct. 25) [Online]. Available: http://www.honeywellmicroelectronics.com/rad-hard-MRAM.php

[4] C. Gopalan, Y. Ma, T. Gallo, J. Wang, E. Runnion, J. Saenz, F. Koushan, P. Blanchard, and S. Hollmer, “Demonstration of Conductive Bridging Random Access Memory (CBRAM) in logic CMOS process,” Solid-State Electron., vol. 58, pp. 54-61, 2011.

[5] Assessment of the potential and maturity of selected emerging research memory technologies, ITRS Workshop [Online], available: http://www.itrs.net/Links/2010ITRS/2010Update/ToPost/ERD_ERM_2010FINALReportMemoryAssessment_ITRS.pdf

[6] Richard Fackenthal, Makoto Kitagawa, Wataru Otsuka, Kirk Prall, Duane Mills,1 Keiichi Tsutsui,4 Jahanshir Javanifard, Kerry Tedrow, Tomohito Tsushima, Yoshiyuki Shibahara, Glen Hush. "A 16Gb

> Submission to IEEE-TNS <

6

ReRAM with 200MB/s Write and 1GB/s Read in 27nm Technology" Proceedings ISSCC 2014

[7] [Online].Available: http://www.eetimes.com/document.asp?doc_id=1327289, “Intel, Micron Launch "Bulk-Switching" ReRAM,” Peter Clarke, 7/28/2015

[8] CBRAM is a registered trademark of Adesto Technologies Corporation [Online]. Available: http://www.adestotech.com/CBRAM

[9] Y. Gonzalez-Velo, H. J. Barnaby, M. N. Kozicki, et al., “Total-ionizing dose effects on the resistance switching characteristics of chalcogenide programmable metallization cells,” IEEE Trans. Nucl. Sci., vol. 60, no. 6, pp. 4563–4569, Dec. 2013.

[10] P. Dandamudi, H. J. Barnaby, M. N. Kozicki, et al., “Total ionizing dose tolerance of the resistance switching of Ag-Ge40S60 based programmable metallization cells,” IEEE Trans. Nucl. Sci., in press.

[11] W. Chen, H. J. Barnaby, M. N. Kozicki, Y. Gonzalez-Velo, R. Fang, K. Holbert, S. Yu, W. Yu, A Study of Gamma-ray Exposure of Cu-SiO2 Programmable Metallization Cells, accepted for publications IEEE Trans. Nucl. Sci., Dec 2015

[12] M. N. Kozicki, M. Mitkova, M. Park, M. Balakrishnan, C. Gopalan, “Information storage using nanoscale electrodeposition of metal in solid electrolytes,” Superlattices and Microstructures, vol. 34, pp. 459-465, December 2003.

[13] M. N. Kozicki, M. Park, M. Mitkova, “Nanoscale memory elements based on solid-state electrolytes,” IEEE Trans. on Nanotechnology, vol. 4, pp. 331-338, May 2005.

[14] I. Valov and M. N. Kozicki, “Cation-based resistance change memory,” J. Phys. D, Appl. Phys., vol. 46, no. 7, p. 074005, Feb. 2013.