You are on page 1of 7

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 66, NO.

5, MAY 2019 2179

Demonstration and Understanding of Nano-RAM


Novel One-Time Programmable
Memory Application
Sheyang Ning , Member, IEEE , and Jia Luo

Abstract — In prior researches, carbon nanotube (CNT)- is used for this conventional high endurance application in
based nano-random access-memory (NRAM) uses reset order to differentiate with the one-time programmable (OTP)
initialization which obtains >1011 high write endurance for application proposed in this paper.
large-time programmable (LTP) application. In this paper, for
the first time, NRAM one-time programmable (OTP) applica- Different from the conventional LTP NRAM, for the first
tion is proposed by using set initialization for data archive. time, NRAM OTP application is demonstrated and analyzed
Specifically, virgin NRAM cells are all in high resistance which uses the same virgin cell as LTP. Specifically, virgin
state (HRS) for storing bit “0.” In contrast, set initialization NRAM cells are all in HRS which can be used to store bit “0.”
uses reversed polarity voltage to obtain low resistance In contrast to LTP NRAM, OTP requires set initialization with
state (LRS) for bit “1.” Furthermore, physical models of
set initialization and retention current degradation are pro- reversed voltage polarity to reduce cell resistance to low resis-
posed for the first time. The current increment during set tance state (LRS) for storing bit “1.” The HRS is more stable
initialization can be attributed to electron tunneling and than LRS on data retention and read disturb because the virgin
CNT deformation. Retention current degradation may be due NRAM cells have been through high temperature in back-
to variation of paralleled CNT contacts in bottleneck zone. end-of-line (BEOL) and plasma etching caused voltage stress
As for OTP performance, median NRAM bit and 1% tail bit
demonstrate more than 1 billion years and 15 years data during manufacture. Extensions of measured retention results
retentions, respectively, on 150 °C. The tail bit activation demonstrate that the median bit and 1% tail bit obtain more
energy is 2.41 eV. Finally, no LRS read disturb is found after than 1 billion years and 15 years data retentions, respectively,
10 s 0.5-V stress on both polarities. The virgin NRAM cells in on 150 °C. No read disturb is found after 10 s 0.5-V stress.
HRS should be more stable than set initialized NRAM cells OTP is widely used on applications such as boot code,
in LRS for both retention time and read disturb.
encryption keys, and configuration parameters for analog,
Index Terms — Carbon nanotube (CNT), data retention, sensor, and display circuitry[5]–[10]. In addition, high capac-
digital archive, emerging memory, error correcting code ity, long retention, or permanent retention data archive in
(ECC), nano-random access memory (NRAM), nonvolatile
memory, one-time programmable (OTP) memory.
specific applications becomes a hot topic [11], [12]. However,
conventional OTP devices may not be suitable for the archive
application. Specifically, fuse and antifuse OTPs have large
I. I NTRODUCTION cell size which limits memory capacity [9], [10]. In addi-
tion, though NAND flash obtains the highest cell density and
C ARBON nanotube (CNT)-based nonvolatile nano-
random access-memory (NRAM) is one of the promis-
ing candidates of emerging memories. It has demonstrated
capacity, it shows limited data retention time and radiation
sensitivity [13], [14]. In contrast, OTP NRAM demonstrates
high performances, such as fast write and read speeds, 1011 potential for data archive application because of its small
high endurance, good data retention, and low program cur- cell size, long data retention time, and radiation-resistant
rent [1]–[4]. Reset initialization is needed on virgin cell characteristic [15].
by applying positive voltage on top electrode (TE) while The rest of this paper is organized as follows. NRAM OTP
grounding the bottom electrode (BE). It is called reset ini- is proposed for the first time in Section II. Measured NRAM
tialization because it uses the same voltage polarity as reset reset and set initialization currents and proposed physical mod-
program. In addition, the term large-time programmable (LTP) els are demonstrated and discussed. Then, Section III shows
OTP data retention performance. A physical model is also pro-
Manuscript received December 30, 2018; revised March 4, 2019 and posed to understand HRS and LRS degradations. In addition,
March 19, 2019; accepted March 20, 2019. Date of publication April 10, OTP NRAM read disturb performance and the comparison
2019; date of current version April 22, 2019. The review of this paper
was arranged by Editor P. Du. (Corresponding author: Sheyang Ning.) with other OTP devices are discussed in this section.
The authors are with the Nantero Inc., Sunnyvale, CA 94085 USA
(e-mail: sning@nantero.com). II. OTP S ET I NITIALIZATION AND P HYSICAL M ODEL
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. In this section, reset and set initialization currents are
Digital Object Identifier 10.1109/TED.2019.2907140 measured on single virgin NRAM cells. Novel NRAM OTP

0018-9383 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
2180 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 66, NO. 5, MAY 2019

Fig. 2. Simplified NRAM process [2].

Fig. 1. Photographs of a planar-type NRAM cell after (a) set program and
(b) reset program [4]. The variation of spacing between CNTs decides
NRAM cell tunneling current. Fig. 3. Comparison of measured set and reset initialization currents by
using (a) graph with linear X- and Y-axes and (b) FN plot.

application is proposed for the first time. Physical models of


reset and set initializations are discussed. initialization uses reversed voltage polarity and provides stable
cell resistance for OTP application. At beginning, all virgin
A. NRAM Photograph and Process NRAM cells are in HRS which can be used for bit “0.”
Fig. 1(a) and (b) shows photographs of a planar-type NRAM Then, set initialization is applied to program NRAM cells to
cell after set and reset programs, respectively [4]. This NRAM LRS for bit “1.” Fig. 3(a) shows the comparison of measured
cell is different from the vertically placed 140-nm single current versus voltage during reset and set initializations on
NRAM cells which are measured in the rest of this paper [3]. two discreate NRAM cells. For set initialization, a current
However, the cell resistance switching mechanism can be compliance is applied to prevent NRAM cell break.
considered the same. Specifically, more CNTs are placed on To understand the physical mechanisms during reset and
one electrode than the other after initialization. A void, or say set initializations, Simmons’s tunneling current equation for
a bottleneck zone with less CNT contacts, can be found in the metal–insulator–metal (MIM) system is showed in (1) [16].
lower middle part between two electrodes. Set program for J is current density. D is the spacing between CNTs. V is the
LTP application applies positive voltage on the electrode with external voltage applied on CNT. ∅ is insulator average
√ barrier
less CNTs, as illustrated in Fig. 1(a). It reduces the spacing height. q is electron charge. In addition, A = 4πβ 2m/ h and
between CNTs so that decreases NRAM cell resistance. On the B = q/(2πβ 2 h) can be considered as constants. In A and B,
other hand, CNT contacts are separated in Fig. 1(b) after h is Planck constant. β is correction coefficient and usually
reset program with reversed voltage polarity. The spacing β ≈ 1 [16]. m is electron mass
between CNTs decides NRAM cell current due to electron B √ √
−AD ∅
tunneling. Moreover, a simplified NRAM process is shown in J= 2
[(∅e − (∅ + q V )e−AD ∅+q V ]. (1)
D
Fig. 2 [2]. In detail, CNT layer is spin-coated between two
metal layers [3]. Conventional dry etching and postetching To simplify this equation, (2) shows an approximation in
cleaning are used for fabricating NRAM cells [3]. The virgin direct tunneling region when low bias voltage is applied
√ and
NRAM cell in HRS has been through standard 425 °C BEOL ∅  q V [16]. C is a constant which equals q 2 2m/ h 2 .
temperature and plasma etching caused voltage stress during Equation (2) shows that the junction is ohmic in direct
manufacture [3]. So, HRS’s data retention and read disturb tunneling region when the spacing D is fixed

should be more stable than LRS after set initialization. √ − AD ∅
3C ∅V β
J= e . (2)
B. Physical Model of Reset Initialization 2D
Conventional NRAM developments and researches focus Furthermore, (3) shows another approximation of (1) in
on high endurance LTP application. It requires reset initial- Fowler–Nordheim (FN) tunneling region by using large bias
ization by applying positive voltage on TE. In contrast, set voltage so that ∅ < q V [16]. C  = 2.2q 3 /(8πh) and
NING AND LUO: NRAM NOVEL OTP MEMORY APPLICATION 2181

Fig. 4. Proposed physical mechanism for set initialization.


C  = 8π 2m/(2.96 ∗ hq) are both constants in the following
equation [16]:
Fig. 5. Validation of the set initialization model which combines electron
√ 3 tunneling and CNT elastic deformation.
V 2 −C  D ∅0
J = C e V . (3)
∅0 D 2
To understand the mechanism of reset initialization,
Fig. 3(b) shows a well-known FN plot by adjusting Fig. 3(a)’s
X- and Y -axes. The reset initialization shows log-like curve in
low-voltage region and negatively sloped curve in high-voltage
region which match with (2) and (3), respectively. This result
can also be found frequently in other MIM solid materials.
So, it suggests that during reset initialization, the spacing
between CNTs does not have obvious variation on direct
tunneling and FN tunneling regions until a higher voltage
Fig. 6. (a) Photograph of 4-Mbit 140-nm NRAM cell array [3], [19].
causes separation. In prior publications, the separation of (b) Array structure and voltage applied for set initialization.
CNTs is attributed to phonon vibration [1]–[4].

C. Physical Model of Set Initialization Fig. 5 shows the validation of (6) by adjusting X- and
Y -axes from Fig. 3(a). Linear curve is expected in low bias
In Fig. 3(b), set initialization has higher current on low region. The result in Fig. 5 shows that this assumption has a
bias which cannot be explained by using direct tunneling reasonable match from measured set initialization current.
or FN tunneling. Naturally, the CNT spacing reduction as
shown in Fig. 1 should also be considered for explanation
III. OTP R ETENTION P ERFORMANCES
because the cell resistance is converted from high resistance
AND P HYSICAL M ODEL
to low resistance after set initialization. A cartoon figure in
Fig. 4 provides a clear demonstration of this phenomenon. The In this section, OTP NRAM data retention performances
occurrence of CNT deformation is on the current bottleneck are measured on NRAM array. Then, a physical model for
because it has the largest voltage drop and electrostatic force. data retention is proposed. Next, read disturb performance is
To analyze this assumption, electrostatic force caused CNT demonstrated and discussed. Finally, OTP NRAM is compared
approaching is calculated in (4). The left side is electrostatic with other devices for data archive application.
force when voltage V is applied on CNTs [17]. It equals to
elastic force due to CNT or CNT network deformation [18]. A. Improvement of NRAM Array Set Initialization
ε is relative permittivity. S is effective CNTs overlap area. Higher voltage and longer pulse width can be used on
K is elastic modulus. D0 is initial CNT spacing before set the set initialization for larger cell current and better data
initialization retention. The strong initialization stress may further reduce
εSV 2 the spacing between CNTs or increase the number of carbon
= −K (D0 − D). (4) atoms in contact. In this subsection, NRAM cell resistance
2D 2
after set initialization is compared to find the best set ini-
To give a qualitative analysis, (5) shows derived relationship
tialization scheme. First, Fig. 6(a) and (b) shows the pho-
between D and V
 tograph of 140-nm NRAM cell array and the schematic of
3
D ∝ V 2. (5) one-transistor one-resistor (1T1R) structure, respectively [19].
High voltage is applied on bitline (BL) and NRAM BE for set
Then, (5) is brought to (2) for finding the I –V correlation initialization. The word line uses higher voltage for passing BL
in low bias region as shown in the following equation: voltage.
√ √ By using this array, Fig. 7(a) and (b) shows the compari-

3 −A
3
V2 ∅
I ∝ Ve β . (6) son of the performance of four set initialization conditions.
2182 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 66, NO. 5, MAY 2019

Fig. 7. (a) Comparison of set initialization conditions. (b) Comparison


of measured median cell current of 2-Kbit per condition after set initial-
ization.

The first one uses single 100-μs pulse with fixed voltage.
The second one reduces the pulsewidth by 10 times. The
third condition separates the stress time into 1000 pulses with
100-ns per pulse. The program voltage V used in these three
conditions equals to the maximum voltage that this process
can tolerate. Finally, the last set initialization reduces the
program voltage by 0.5-V. For each condition, set initialization
is applied on 2-Kbit. Then, the median cell’s read current
is compared in Fig. 7(b). Though the conditions 1 and 3
have same program voltage and total stress time, multiple
pulse programs in condition 3 obtain the highest cell current.
This can be attributed to prior published NRAM characteristic Fig. 8. Measured cell current distribution when baking at (a) 85 °C
and (b) 150 °C.
that set and reset programs have higher success rate on
leading edge of program pulse [2]. In addition, the comparison
between conditions 3 and 4 also shows that high voltage is the
key for NRAM program. Finally, higher cell current and longer extension of current degradation reaches the sensing criteria.
data retention can be expected by further increasing the set Furthermore, the red curves are the median cell current of the
initialization voltage. However, higher voltage than V is pro- 1% tail bits in three dies. So, the red curve is considered as
hibited in our test, due to transistor limitation in control circuit. cell current degradation at 0.5% tail bit. From these figures,
both median current and tail bit current have linear degradation
when retention time is in log scale.
B. Measured Data Retention Performances The data retention times for median bit and tail bit are
By using the best set initialization scheme, Fig. 8(a) and (b) then compared in Fig. 10(a) and (b), respectively. The raw
shows measured 4-Kbit OTP cell current distribution percentile data from Fig. 9(a)–(c) are used in these analyses. Each
when applying retention on 85 °C and 150 °C, respectively. LRS curve shows a combined result from three dies with
In Fig. 8(a) and (b), HRS resistance does not move because the 4-Kbit per die. In addition, for each HRS curve, 2-Kbit are
virgin NRAM cell has been through much higher temperature measured on virgin NRAM die. In both figures, LRS cell
in manufacture. On LRS side, NRAM cell current degradation current degradation shows larger slope when applying data
saturates quickly during 150 °C baking. By using low-density- retention on higher temperature. In Fig. 10(a), HRS median
parity-check (LDPC) error correcting code (ECC), more than bit does not find current variation when baking on 85 °C
1% raw bit error rate (RBER) can be corrected by using or 150 °C. In addition, the extended 150 °C LRS and HRS
0.89 high code rate and soft decoding [20]. So, this paper curves still keep large sensing margin even after 1 billion years
uses 1% RBER margin and evenly split to 0.5% for LRS and retention. Furthermore, Fig. 10(b) shows 1% tail bit current
0.5% for HRS in the following analyses. degradations including 0.5% for HRS and 0.5% for LRS. The
In this section, more efforts and analyses are applied on LRS sensing criteria are decided by the cross point of 150 °C LRS
side because it shows faster retention degradation. In specific, and HRS extensions. More than 15 years data retention time
Fig. 9(a)–(c) shows 10 days data retention on 85 °C, 125 °C, can be obtained on 150 °C. In addition, the retention times on
and 150 °C, respectively. In total, nine NRAM dies (parts) and 125 °C and 85 °C are decided by LRS current degradation.
4-Kbit per die are measured. In each figure, the three black Though no margin is assigned in Fig. 10(b), the assumption
curves with higher current show the median cell current of the and calculation are practical and agree with industry standard
three dies with 4-Kbit measured per die. The gray curves in the data retention time projection. In detail, normally the margin
background are the 1% tail bits of the three dies. The definition is reserved after program verify for compensating margin
of tail bit is having the shortest data retention time when the loss due to program disturb, read disturb, data retention
NING AND LUO: NRAM NOVEL OTP MEMORY APPLICATION 2183

Fig. 9. Measured LRS NRAM cell current degradation during 10 days baking by using (a) 85 °C, (b) 125 °C, and (c) 150 °C. In all figures, both
median bit and 0.5% tail-bit show linear current degradation when data retention time is in log scale.

So, the sensing circuit baking caused margin loss is also


included in this figure.
Besides using LDPC, the data retention time is also cal-
culated by using Bose–Chaudhuri–Hocquenghen (BCH) ECC
which is another popular ECC scheme for nonvolatile memory.
It has smaller circuit size but the correctable RBER is reduced
to about 0.4% [20]. At 150 °C, retention time is largely
reduced from more than 15 years by using LDPC to 1 year
by using BCH. Similarly, at 85 °C, retention time is reduced
from 2 × 106 years to 2 × 105 years by using BCH.
Arrhenius equation (7) is commonly used to calculate data
retention times and active energy on resistive RAM (RRAM),
phase change RAM (PRAM), and conductive bridge RAM
(CBRAM) [21]–[24]. For example, it is suitable for RRAM
cell current degradation which is due to oxygen vacancy
diffusion [21], [24]. Similarly, NRAM retention degradation
can be attributed to displacement and diffusion of CNT–CNT
contacts. So, the Arrhenius equation is also used in this
paper for calculating OTP NRAM retention degradation and
activation energy

t = t0 e Ea / k B T . (7)

In (7), T is absolute temperature. t is data retention time


on temperature T . t0 is retention time on infinite temperature.
k B is Boltzmann’s constant. E a is activation energy for 1% tail
bit data retention. By using data retention time on 85 °C and
150 °C, the calculated activation energy is 2.41eV. Further-
Fig. 10. Measured (a) median cell current and (b) 1% tail-bit current more, extended curve in Fig. 11 shows more than 1020 seconds
(0.5% HRS and 0.5% LRS) degradations on different temperatures. For retention time on 25 °C.
each temperature, data from three parts are combined.

C. Data Retention Physical Model


degradation, and sensing noise. In NRAM OTP application, According to Ohm’s law, the bottleneck of current flow on
there is no program disturb concern due to the 1T1R structure, the CNT sparse region should make key contribution to cell
and the user data are written only once. The read disturb current variation during retention. Based on the photographs
is also not a concern as show in the following analyses. in Fig. 1, the bottleneck is made from paralleled CNT contact
Furthermore, the data retention caused margin degradation is pairs. Then, by considering measured results in Figs. 9 and 10,
closely monitored in Fig. 10(b). Finally, the sensing noise a physical model of NRAM OTP data retention is proposed
is already included in Fig. 10 because the data is read out as below.
from our on-die sense amplifier. In addition, during the data First, the NRAM LRS current reduction and HRS current
retention test, sensing circuit is also baked with memory cells. growth during data retention clearly suggest an equilibrium
2184 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 66, NO. 5, MAY 2019

Fig. 12. Comparison of NRAM cell current distributions before and after
0.5-V, 10-s read stress by using (a) TE read and (b) BE read.

Fig. 11. Arrhenius plot of data retention time. TABLE I


C OMPARISON OF M EMORIES FOR D IGITAL A RCHIVE A PPLICATION
state with moderate number of paralleled CNT contact pairs
or contact atoms in the bottleneck region. Equation (8)
demonstrates measured linear-log relation of current and reten-
tion time for HRS and LRS approaching to the equilibrium
state. In (8), A is the slope of retention curve as shown
in Figs. 9 and 10
I ∝ Aln(t). (8)
Then, (9) calculates the current variation rate dI/dt in each
NRAM cell by adding or removing part of paralleled CNT
contacts which have resistance R. The result shows that it
changes the NRAM cell resistance by 1/R. In (9), R0 is
initial cell resistance before data retention
 
dI 1 1 V
=V − = . (9)
dt R0 //R R0 R
E. Comparison With Other OTP Devices
Finally, the retention current degradation as shown in (8)
matches with measured result. It can be derived by solving Table I shows the comparison of NRAM performances with
the integration in (9) if V /R can be replaced by V × A/t other nonvolatile memories for OTP data archive application.
as shown in (10). The n in (10) is the number of reduced or In detail, antifuse demonstrates stable data retention after
increased CNT contacts in bottleneck zone which cause 1/R 150 °C, 450-h without degradation [26]. In addition, EEPROM
resistance variation on NRAM cell can obtain 10-year data retention time on 150 °C [27]. How-
ever, both antifuse and EEPROM have limited capacity due
1 A
= ∝ n. (10) to large cell size. For example, at 10-nm technology node,
R t 2T antifuse needs 144-nm × 192-nm cell size [28]. On the
In conclusion, the model shows that during data retention, other hand, 3-D NAND flash obtains the highest capacity due to
the increase or decrease of paralleled CNT contacts n in bot- 3-D array structure and multiple bits per cell. However, it has
tleneck zone is inversely proportional to retention time t. The limited data retention time and radiation sensitivity [13], [14].
reduction of LRS current during retention can be considered Furthermore, emerging memories such as RRAM, PRAM, and
as recover of the CNT deformation during set initialization. NRAM have potential of high array capacity because they
On the other hand, the increase of HRS current may be also all support 3-D array structure, and have less limitation for
due to unbalanced CNT net, from mechanical stress or plasma cell etching [29]–[30]. For example, Rueckes [3] demonstrated
etching caused voltage stress on virgin NRAM cell. 22-nm NRAM cell resistance switching. Among these emerg-
ing memories, OTP NRAM shows the best data retention
D. Read Disturb
performance. In addition, NRAM also has advantage of
Fig. 12(a) and (b) shows OTP read disturb performances by radiation-resistant characteristic [15].
applying read voltage on TE and BE, respectively. LRS read
disturb is demonstrated because it is less stable than HRS.
IV. C ONCLUSION
Specifically, one second 0.5-V stress is applied 10 times on
three NRAM dies. No obvious cell resistance shift is found in A novel NRAM OTP application is proposed for the first
both figures. From prior publication, NRAM array requires time. Virgin NRAM cells in HRS are used for data “0,”
only 13.5-ns access time [25]. So, the applied total 10-s whereas set initialization is needed to program NRAM cells
read stress shows no read disturb concern on NRAM OTP to LRS for data “1.” Two physical models are also proposed
application. for understanding set initialization current and LRS, HRS
NING AND LUO: NRAM NOVEL OTP MEMORY APPLICATION 2185

retention current degradation. OTP NRAM is suitable for data [14] D. N. Nguyen, S. M. Guertin, and J. D. Patterson, “Radiation tests
archive application because it has long data retention on high on 2Gb NAND flash memories,” in Proc. IEEE Radiat. Effects Data
Workshop, Jul. 2006, pp. 121–125. doi: 10.1109/REDW.2006.295479.
temperature, potential of high capacity, and radiation-resistant [15] (Nov. 18, 2009). Lockheed Martin Tests Carbon Nanotube-Based
characteristic. Memory Devices On NASA Shuttle Mission. [Online]. Available:
https://news.lockheedmartin.com/2009-11-18-Lockheed-Martin-Tests-
ACKNOWLEDGMENT Carbon-Nanotube-Based-Memory-Devices-on-NASA-Shuttle-Mission,
[16] J. G. Simmons, “Generalized formula for the electric tunnel effect
The authors would like to thank W. Wong, M. Kwan, between similar electrodes separated by a thin insulating film,” J. Appl.
M. Ramsbey, T. Gallagher, N. Leong, and L. Cleveland for Phys., vol. 34, no. 6, pp. 1793–1803, Jul. 1963. doi: 10.1063/1.1702682.
[17] A. Hajnayeb, S. E. Khadem, and M. Zamanian, “Thermoelastic damp-
their supports and discussions. ing of a double-walled carbon nanotube under electrostatic force,”
IET Micro Nano Lett., vol. 6, no. 8, pp. 698–703, Aug. 2011.
R EFERENCES doi: 10.1049/mnl.2011.0193.
[1] G. Rosendale et al., “Storage element scaling impact on CNT memory [18] Deformation (Engineering). Accessed: Dec. 20, 2018. [Online].
retention and ON/OFF window,” in Proc. IEEE 6th Int. Memory Work- Available: https://en.wikipedia.org/wiki/Deformation_(engineering)#
shop (IMW), May 2014, pp. 1–3. doi: 10.1109/IMW.2014.6849391. Elastic_deformation
[2] S. Ning et al., “23% faster program and 40% energy reduction of [19] S. Ning et al., “Carbon nanotube memory cell array program error
carbon nanotube non-volatile memory with over 1011 endurance,” in analysis and tradeoff between reset voltage and verify pulses,”
Proc. Symp. VLSI Technol. Dig. Tech. Papers, Jun. 2014, pp. 1–2. Jpn. J. Appl. Phys., vol. 55, pp. 04EE011–04EE016, Feb. 2016.
doi: 10.1109/VLSIT.2014.6894384. doi: 10.7567/JJAP.55.04EE01.
[3] T. Rueckes, “High Density, High Reliability Carbon Nanotube NRAM,” [20] K. Zhao, W. Zhao, H. Sun, T. Zhang, X. Zhang, and N. Zheng, “LDPC-
in Proc. Flash Memory Summit presentation, Aug. 2011, pp. 1–12. in-SSD: Making advanced error correction codes work effectively in
[4] S. Ning et al., “Investigation and improvement of verify-program solid state drives,” in Proc. 11th USENIX Conf. File Storage Technol.
in carbon nanotube-based nonvolatile memory,” IEEE Trans. (FAST), 2013, pp. 1–14.
Electron Devices, vol. 62, no. 9, pp. 2837–2844, Sep. 2015. [21] Y. Koo, S. Ambrogio, J. Woo, J. Song, D. Ielmini, and H. Hwang,
doi: 10.1109/TED.2015.2450219. “Accelerated retention test method by controlling ion migration barrier
[5] P.-R. Cheng, C.-S. Yang, M.-Y. Hsu, C. J. Lin, and Y.-C. King, “Variable- of resistive random access memory,” IEEE Electron Device Lett., vol. 36,
length gateless transistor for analog one-time-programmable memory no. 3, pp. 238–240, Mar. 2015. doi: 10.1109/LED.2015.2394302.
applications,” in Proc. Int. Symp. VLSI Technol. Syst. Appl. (VLSI-TSA), [22] J. R. Jameson et al., “Conductive-bridge memory (CBRAM) with
Apr. 2016, pp. 1–2. doi: 10.1109/VLSI-TSA.2016.7480485. excellent high-temperature retention,” IEDM Tech. Dig., Dec. 2013,
[6] T. He, F. Zhang, S. Bhunia, and P. X.-L. Feng, “Silicon car- pp. 3011–3014. doi: 10.1109/IEDM.2013.6724721.
bide (SiC) nanoelectromechanical antifuse for ultralow-power one- [23] G. Navarro et al., “Trade-off between SET and data reten-
time-programmable (OTP) FPGA interconnects,” IEEE J. Elec- tion performance thanks to innovative materials for phase-change
tron Devices Soc., vol. 3, no. 4, pp. 323–335, Jul. 2015. memory,” in IEDM Tech. Dig., Dec. 2013, pp. 2151–2154. doi:
doi: 10.1109/JEDS.2015.2421301. 10.1109/IEDM.2013.6724678.
[7] R. Barsatan, T. Y. Man, and M. Chan, “A zero-mask one-time program- [24] Z. Wei et al., “Demonstration of high-density ReRAM ensuring
mable memory array for RFID applications,” in Proc. IEEE Int. Symp. 10-year retention at 85 °C based on a newly developed relia-
Circuits Syst., May 2006, p. 4. doi: 10.1109/ISCAS.2006.1692750. bility model,” in IEDM Tech. Dig., Dec. 2011, pp. 3141–3144.
[8] T.-Y. Lee, C.-C. Chiu, Y.-C. Liu, C.-C. Liu, Y.-C. King, and doi: 10.1109/IEDM.2011.6131650.
C.-J. Lin, “A new embedded one-time-programmable MNOS mem- [25] B. Gervasi. (2018). Architecture for Carbon Nanotube Based
ory fully compatible to LTPS fabrication for system-on-panel (SOP) Memory (NRAM). [Online]. Available: http://www.hotchips.org/hc30/
applications,” IEEE Electron Device Lett., vol. 29, no. 8, pp. 906–908, 2conf/2.04_Nantero_20180818_hotchips_gervasi_nram_presentation.pdf
Aug. 2008. doi: 10.1109/LED.2008.2000831. [26] W. Y. Hsiao et al., “A high density Twin-Gate OTP cell in pure 28 nm
[9] H.-K. Cha et al., “A 32-KB standard CMOS antifuse one-time CMOS process,” in Proc. Int. Symp. VLSI Technol. Syst. Appl. (VLSI-
programmable ROM embedded in a 16-bit microcontroller,” IEEE TSA), Apr. 2014, pp. 1–2. doi: 10.1109/VLSI-TSA.2014.6839664.
J. Solid-State Circuits, vol. 41, no. 9, pp. 2115–2124, Sep. 2006. [27] G. Schatzberger, F. P. Leisenberger, P. Sarson, and A. Wiesner, “High
doi: 10.1109/JSSC.2006.880603. efficient low cost EEPROM screening method in combination with an
[10] M. Shi et al., “Zero-mask contact fuse for one-time-programmable area optimized byte replacement strategy which enables high reliability
memory in standard CMOS processes,” IEEE Electron Device Lett., EEPROMs,” in Proc. IEEE 36th VLSI Test Symp. (VTS), Apr. 2018,
vol. 32, no. 7, pp. 955–957, Jul. 2011. doi: 10.1109/LED.2011.2147754. pp. 1–6. doi: 10.1109/VTS.2018.8368631.
[11] T. Hilbel, B. D. Brown, J. de Bie, R. L. Lux, and H. A. Katus, [28] S.-Y. Chou, Y.-S. Chen, J.-H. Chang, Y.-D. Chih, and T.-Y. J. Chang,
“Innovation and advantage of the DICOM ECG standard for view- “A 10 nm 32Kb low-voltage logic-compatible anti-fuse one-time-
ing, interchange and permanent archiving of the diagnostic electro- programmable memory with anti-tampering sensing scheme,” in IEEE
cardiogram,” in Proc. Comput. Cardiol., Sep./Oct. 2007, pp. 633–636. Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. 2017,
doi: 10.1109/CIC.2007.4745565. pp. 200–201. doi: 10.1109/ISSCC.2017.7870330.
[12] S. Alam, M. Kelly, and M. L. Nelson, “Interplanetary wayback: The per- [29] D. L. Lewis and H.-H. S. Lee, “Architectural evaluation of 3D stacked
manent Web archive,” in Proc. IEEE/ACM Joint Conf. Digit. Libraries RRAM caches,” in Proc. IEEE Int. Conf. 3D Syst. Integr., Sep. 2009,
(JCDL), Jun. 2016, pp. 273–274. doi: 10.1145/2910896.2925467. pp. 1–4. doi: 10.1109/3DIC.2009.5306582.
[13] K. Mizoguchi, T. Takahashi, S. Aritome, and K. Takeuchi, “Data- [30] W. Zhang and T. Li, “Exploring phase change memory and 3D
retention characteristics comparison of 2D and 3D TLC NAND flash die-stacking for power/thermal friendly, fast and durable memory archi-
memories,” in Proc. IEEE Int. Memory Workshop (IMW), May 2017, tectures,” in Proc. 18th Int. Conf. Parallel Archit. Compilation Techn.,
pp. 1–4. doi: 10.1109/IMW.2017.7939077. Sep. 2009, pp. 101–112. doi: 10.1109/PACT.2009.30.

You might also like