서지주요정보
Unified random access memory with nanowire and energy band engineering = 에너지 밴드 엔지니어링 및 나노와이어 기반의 융합메모리
서명 / 저자 Unified random access memory with nanowire and energy band engineering = 에너지 밴드 엔지니어링 및 나노와이어 기반의 융합메모리 / Jin-Woo Han.
발행사항 [대전 : 한국과학기술원, 2010].
Online Access 원문보기 원문인쇄

소장정보

등록번호

8021089

소장위치/청구기호

학술문화관(문화관)B1층 보존서고

DEE 10043

휴대폰 전송

도서상태

이용가능(대출불가)

사유안내

반납예정일

리뷰정보

초록정보

It’s time for a paradigm shift in a silicon-based memory. $\It{‘Hwang’s Law’}$, the memory capacity doubling every year, has been the basis of 8-year technological and economic revolution by shrinking of transistors. Whether this size scaling has enabled the exponential growth in a semiconductor industry, it will eventually come to an end in the near future. The focus in the memory will have to shift to devices that are not just increasingly scalable but multi-functions supportable. As a technological breakthrough that can overcome the saturation in the revenue obtained from $\It{‘scaling’}$, a novel type of fusion memory is presented. Combined with a charge-trapping gate dielectric in a floating body transistor, functions of a non-volatile flash memory and high-speed capacitorless 1T-DRAM are integrated in a single memory transistor. The memory named $\underbar{U}nified \underbar{R}andom \underbar{A}ccess \underbar{M}emory$ (URAM) could radically improve the performance of embedded system without shrinking memory cells, and the paradigm shift from $\It{‘scaling’}$ to $\It{‘multi-function’}$ will create new value and continue the evolution of silicon memory technology. In the end, URAM will in turn become the cornerstone of new silicon-based memory that operates versatile functions of high-speed mode and non-volatile mode according to the end user’s demand.

반도체 산업의 발전은 ‘무어의 법칙’에 따라 작게 만들면 작게 만들수록 성능 증가, 집적도 증가, 단가 감소의 원칙을 따라왔다. 하지만 근래에 소자의 선폭이 50nm 미만으로 내려가면서 물리적인 한계와 공정기술의 어려움 때문에 소자의 소형화 속도가 점점 늦어지고 있다. 소형화의 속도가 점점 늦어지고 있다. 이것은 반도체 산업의 대 원칙인 ‘무어의 법칙’의 종말도 머지 않았음을 암시한다. 이것을 증명하는 한 예로 물리적인 한계라고 일컬어 지는 3nm 근방의 소자도 연구단계에서 속속 등장하고 있다. 반도체 기술의 발전을 계속 이어가고 새로운 부가가치를 창출하기 위해서는 이제 더 이상 ‘소형화’에 의존하지 않고 새로운 개념의 소자 개발이 필요하다. 이러한 요구에 대응하기 위해 단일 소자에 복합 기능을 부여하는 ‘다기능 메모리 소자’ 개념을 제안하였다. 메모리는 주로 PC에서 사용되는 DRAM과 디지털 기기(MP3, 디지털카메라)에서 사용되는 Flash로 나뉜다. DRAM은 매우 빠른 처리속도를 갖지만 전원을 끄면 데이터가 사라진다. 반대로 Flash는 전원이 없어도 정보가 유지되는 비 휘발성이지만 처리속도가 매우 느리다. 따라서 각각의 장점에 따라 시장이 양분되어 있다. 본 연구는 각각의 장점을 한 소자에서 구현되는 통합메모리에 대한 것이다. 두 가지 메모리가 서로 동작전압이 상이한 것에 착안하여, 동작전압에 따라 DRAM동작을 하거나 Flash메모리 동작을 하는 Unified-RAM을 제안한다. 특히 DRAM을 위하여 일반적인 1T/1C DRAM이 아닌 전하를 커패시터가 아닌 SOI 소자의 부유몸체에 저장하는 Capacitorless 1T-DRAM을 사용하였다. 부위몸체 기판에 소자를 제작하여 1T-DRAM 동작을 가능하게 하고, 게이트 절연막에 O/N/O를 사용하여 비휘발성 메모리 동작을 구현하였다. 일반적인 부유 몸체는 SOI 기판을 사용해 왔다. 하지만 SOI 기판의 비싼 가격과 열전달문제 등을 해결하기 위해, 값이 싼 벌크 기판에서 전하를 저장할 수 있는 ‘에너지밴드엔지니어링’ 기판을 제작하였다. Si와 SiC의 가전도대의 오프셋을 이용하여 에너지 장벽 내부에 전하를 저장하거나, Si와 SiGe의 가전도대의 오프셋을 이용하여 에너지 우물 내부에 전하를 저장할 수 있는 기판을 제공하였다. 이러한 두 가지 기능을 수행할 때 두 동작모드의 상호 간섭을 줄이기 위해 구조적으로는 게이트-드레인 언더랩 구조를 제안하였고, 동작방법으로는 충돌이온화를 이용한 충전이 아닌 게이트유도드레인전류 (GIDL)을 사용한 방법으로 간섭이 최소화 됨을 확인하였다. DRAM 읽기 동작 시, 채널에 전하가 저장된 상태를 교란시키지 않는 범위 내에서 MOS읽기 방식을 하게 되고, 이것은 채널의 최소 부피를 요구하므로 소자 소형화를 막는 역할을 한다. 본 연구는 MOS읽기 방식 대신 기생 BJT읽기 방식을 제안하여 부분공핍소자에서 뿐만 아니라 완전공핍소자에서도 DRAM동작을 확인하였고, 이 방법이 소자의 궁극적인 구조라고 할 수 있는 전면게이트 구조에서도 적용된다는 것을 실험적으로 밝혔다. 소자의 소형화를 통한 메모리용량증대는 서두에서 언급한 바와 같이 한계를 지니고 있어, 그 대안으로 3차원 적층방식이 주목을 끌고 있다. 이 시점에서 여러 가지 3차원 적층방식 중 하나인 폴리실리콘채널 적층 방법을 URAM에 적용해 보았다. 폴리실리콘채널을 사용해도 Flash 및 DRAM 동작이 가능함을 보였다. 마지막으로, 기존의 고체 상태의 게이트 절연막을 기체상태로 만들고 채널을 공기중에 부양시켜 채널이 기계적인 움직임을 하는 전기-기계소자를 제작하였다. 공기를 통한 게이트 누설 전류는 이상적인 경우에 영이 되기 때문에 대기전력을 매우 낮출 수 있는 소자로써 사용될 수 있다. 전기-기계소자는 채널의 기계적인 위치에 따라 메모리의 상태가 결정지어 지게 되고 전기적인 특성에 따라 기계적인 위치를 파악할 수 있다. 본 연구는 근 미래에 사용될 수 있는 실용적인 연구로써, 미리의 반도체 산업이 경쟁적으로 유지되고 소자의 소형화를 통한 부가가치를 창출에서 벗어나 새로운 기능을 부여한 것으로 부가가치를 창출하는데 도움을 줄 수 있다. 뿐만 아니라 연구자 에게는 연구 패러다임은 더 이상 소자 소형화에 한계를 두지 않고 다기능화에 있다는 것을 암시하는 의미를 부여한다.

서지기타정보

서지기타정보
청구기호 {DEE 10043
형태사항 iv, 141 p. : 삽화 ; 26 cm
언어 영어
일반주기 저자명의 한글표기 : 한진우
지도교수의 영문표기 : Yang-Kyu Choi
지도교수의 한글표기 : 최양규
학위논문 학위논문(박사) - 한국과학기술원 : 전기및전자공학과,
서지주기 Includes references.
QR CODE

책소개

전체보기

목차

전체보기

이 주제의 인기대출도서

Operational principal ofthe SONOS flash memory

Operational principal ofthe capacitorless IT-DRAM.

Operational bias domain for the flash and the capacitorless IT-DRAM.

The operational sequence of the URAM. Before the high speed mode is activated, NVM should be initialized in order to provide uniform threshold voltage distribution.

Various templates for the floating body effects and energy band diagram for the excess hole storage. The SOI, SOC, and SON are potential barrier type floating substrate, and the SOG is potential well type substrate. Unlike the SOI and SOC, the SON and SOG confines excess hole along the horizontal direction with trench oxide.

Cross-sectional TEM images for p-type MOS capacitors. (a) Sii.yCy on Si

Capacitance-voltage characteristics for the evaluation of energy band-offset.

Energy band diagram for weak and strong accumulation conditions. (a) Si」. yCy on Si substrate and (b) Si/SijxGe, on Si substrate.

(a) Doping profile from SIMS and (b) energy band diagram from simulation

Two types ofthe NOR configuration. (a) the shared source structure uses one contact for two cell, which can enhance layout efficiency, but (b) the divided source structure require one contact for each cell.

Schematics ofneighboring cells cutalong the a-b/c-d directions at the figure 3.3.1. Whereas the SOI and SOC can use the shared source line, the SON and SOG should utilize the divided source line structure.

Process flow ofthe URAM. After the floating substrates are prepared, the subsequent flowis identical.

Bird eye's view ofthe SOI URAM, and cross sectional view ofthe four types ofURAM.

Drain current - gate voltage characteristics for various types ofURAMs. The superior device properties are attributed to the three dimensional device structures.

Device dimensions and parameters for URAMs.

Drain current - drain voltage characteristics for various types of URAMs. The kink points ensure a floating bodyeffect.

Program/erase transient characteristics ofSOI URAM. (a) program transient characteristics. (b) erase transient characteristics.

(a) Retention and (b) endurance characteristics for SOI URAM

Summary ofP/E VT window and AVT after 10 years data retention and 107 P/E cycles for various templates.

Customized measurement system and the operational pulse waveform.

Images of(a) conventional FinFET SONOS and (b) SOI URAM.

Contour of body potential from the device simulation. (a) conventional FinFET SONOS and (b) proposed SOI URAM. In the proposed structure, the lower body potential at the partially depleted zone allows more hole accumulation.

Source current for the capacitorless IT-DRAM. Whereas the two data states are clearly identified according to hole accumulation in the SOI URAM, the source current difference is relativelysmall in the conventional FinFET SONOS.

Simulation profile of the hole concentration biased at hole condition after impact ionization. The profile is cut-plane along the source/drain direction at the center of the fin. (a) conventional FinFET SONOS. (b) SON URAM. In contrast with the case ofthe conventional FinFET, holes are accumulated in the body region in the SON URAM.

Measured source current for the capacitorless IT-DRAM. Whereas two data states are clearly distinguished according to the existence of the holes in the body, the difference between two states is increased as a small amount ofpositive voltage is applied to the substrate. However, as forward p-n junction diode turns on, the capacitorless IT- DRAM is not working.

Measured source current for capacitorless IT-DRAM operation. The small substrate voltage raises the sensing current window.

Simulation profile of the hole concentration biased at hole condition after impactionization. The profileis cut-plane along the width direction at the center ofthe fin. (a) conventional FinFET SONOS. (b) SOC URAM. In contras with the case of the conventional FinFET, holes are accumulated in the body reglon in the SOC URAM.

Energy band offset for different Ge content X. Estimated valence band offset from calibrated C-Vbysimulation, and theoretical data reported in [4.6]-[4.7] are plotted.

(a) Simulation results ofhole concentration biased at hold condition after impact ionization. The excess hole are stored inside the potential well, and (b) the stored hole concentration for various Ge content. The stored hole concentration starts to be saturated over the valence band offset of0.24eV.

Capacitorless IT-DRAM characteristics for different Ge content X. x=0.5 exhibits a wider sensing window but shorter retention time.

(a) TEM images of x=0.4 and 0.3 and (b) schematics for retention degradation mechanisms. The defects reduce the data retention at the programmed state due to the recombination, and a deeper potential well degrade the data retention at the erased state due to hole-to-hole repulsion and its diffusion mechanism.

Summary ofthe features and performance of various URAMs.

P/E characteristics bythe impact ionization method for different program voltages at the drain. The sensing window is reduced at a high drain voltage as the stress timeincreases due to hot electron injection into the nitride traplayer ofO/N/O.

Schematics of excess hole generation method for capacitorless IT-DRAM. (a) impact ionization. (b) band-to-band tunneling or GIDL. Whereas created electrons can be injected into the gate dielectric, created holes are restricted for injection due to thelarge effective mass and high valence band barrier.

P/E characteristics by the GIDL method. An improved sensing current window is observed, thus supporting that the GIDL is an effective mean for the soft- programming free method.

Threshold voltage versus stress time. Threshold voltage is an indicator to estimate the trapped charges in the nitride. GIDLdoes not shift the threshold voltage, while the impact ionization condition does.

Simulation result ofsubstrate current and drain currentint poly-Si versus p + poly-Si gate on a p-type body. Due to the inherent work function difference between p + gateandn drain,thep gate shows higher band-to-band tunneling current.

P/E characteristics for an+ poly-Si versus p poly-Si gate on a p-type body. + Considering that the GIDL current ofthep poly-Si gate is higher than that ofthen poly- Si gate at a given programming voltage, programming efficiency can be improved in the capacitorless IT-DRAM with p poly-Si gate.

Schematic images of(a) gate-to-S/D overlap and (b) gate-to-S/D underlap structure. (c) TEM images of gate-to-S/D underlap devices. The body thickness is 50-nm, the gatelength and width are 110-nm and 350-nm respectively.

Flash memory characteristics. Even though the underlap structure would degrade the P/E efficiency, and acceptable VT window of4.3Vis achieved.

Measurement results of IT-DRAM. The gate-to-S/D underlap structure shows wider sensing current window than the overlap structure due to the reduced electric field and the enlarged volume for hole storage.

Measurement results of capacitorless IT-DRAM. The gate-to-S/D underlap structure shows wider sensing current window than the overlap structure due to the reduced electric field and the enlarged volume for hole storage.

Simulation results ofthe capacitorless IT-DRAM with different spacers. The sensing window of underlap structure is higher than that of overlap structure. And the sensing window of underlap structure is further increased as the high-k offset spacer is employed.

Double-sweep drain current versus the gate voltage characteristics for an SOI FinFET. The fabricated FinFET has a fin width of 30 nm, a gate length of 180 nm, an + O/N/O thickness of3/8/4 nm,anda fin height of110 nm. The device has an n doped poly- Si gate and an undoped body. At VD = 1.8 V, the FinFET shows normal transfer characteristics. AtVp=2.2V, the parasitic BJT begins to work, and the hyst

The operational biases ofthe capacitorless IT-DRAM.

Measured result of the capacitorless IT-DRAM characteristics read by a conventional and the parasitic BJT method. The operational conditions are summarized in Table 6.2.1. In the conventional method, the sensing current window is gradually reduced with time due to a recombination process. In the parasitic BJT method, the source current remains constant with the read time because the BJTread condit

Threshold voltage and sensing current window versus the operation cycles. Threshold voltage shift and sensing current window degradation are found to be negligible, which guarantees very stable URAM operation without soft-programming problem.

Comparison of the different operation methods for the capacitorless 1T- DRAM. The BJT read method shows superior performance and reliability, which leads to the best condition for URAM operation.

Evolution scenario ofthe silicon based MOSFET. The ultimate structure is the gate-all-around nanowire FET.

Scaling length in devices with different geometries

(a) Process flow ofGAA URAM, bird's eye view of(b) the nanowire after the trimming process and (b) final structure. (d) cross-sectional view cut along X-X direction in (c), and (e) magnified imageofgate dielectric stack.

Drain current versus gate voltage characteristics with respect to the various programming conditions for GAA structure (dot) and PD FinFET (line). The pulse width of80 nsec is used for program. Despite of the thick tunneling oxide at GAA structure, the GAA structure exhibits higher programming efficiency. This is attributed the steepened electric field at the tunneling oxide ofGAA structure.

Data retention characteristics at 125'C for GAA structure (square) and PD FinFET (circle). GAA structure shows enhanced charge stability relative to the PD FinFET due to the thick tunneling oxide.

Cyclic endurance comparison between GAA structure (square) and PD FinFET (circle). For both structure, no degradation is monitored in cyclic endurance characteristics.

Measured drain current versus gate voltage characteristics. The data shows the single transistor latch effects in the form ofthe steep slope and hysteresis. The bistable hysteresis can be implemented for the single transistor memory.

The optimized operational conditions and timing diagram for the capacitorless IT-DRAM operation ofGAA URAM. The operation is based on the parasitic BJT read method.

The timing diagram for repeated hold and read for bit '0' and '1', With negative gate voltage, the data state is sustained since the positive base charges are stored in the floating base. And the data states are recovered by applying the drain read voltage or latch-up voltage.

Schematic of the device structure and operational modes in the separated double-gate unified-RAM (URAM). The front- and back-channel are assigned for SONOS flash memory and capacitorless IT-DRAM, respectively. This spatial separation minimizes the soft-programming in the front O/N/O and allows the capacitorless IT-DRAM function regardless ofthe NVM data state.

Operational sequence ofURAM. (a) Tied double-gate URAM [1]-[4] and (b) separated double-gate URAM. In the tied double-gate URAM, an initialization step is required to set the VT state of the capacitorless IT-DRAM and a verification and re- initialization loop is necessary due to the soft-programming issue. In contrast, the separated double-gate URAM eliminates the need for this initialization step

Transmission electron microscopyimage ofthe TFT-based URAM. Aheavily doped back-gate and a back-gate dielectric of6 nm are used. The thickness ofthe undoped poly-Si channel is 20nm, and the thickness ofthe front O/N/O layer is 5/8/7 nm. The gate and spacer length are 90 nm and 40 nm, respectively.

Front- or back-gate voltage (VFG or VBG) versus the drain current (Ip) characteristics The front-channel shows a higher threshold voltage(VT), a lower on-current (Ioo), higher subthreshold slope (SS), anda higher drain-induced barrier lowering (DIBL) a compared to the back-channel because the front-channel utilizes the thick O/N/O dielectric On the other hand, the back-channel shows a higher gate-

Output characteristic. The kink pointin the Ip-VD plot assures the generation of holes and their accumulation, which implies the potential for the capacitorless IT- DRAM.

Transient characteristics for the NVM mode. (a) Programming and (b) erasing. The hot-carrier injection mechanism is utilized for program/erase operations. The P/E conditions at VD.PGM =4.5V/ VFGPGM =10V with CPGM-10jusec for programming and VD.PGM =5.5V/ VFGPPM =-11V with rpaw=300usec for erasing exhibit a threshold voltage window of4V.

(a) Endurance and (b) retention. Atroom temperature, both the retention and endurance characteristics show a negligible amount ofdata loss. At125C,the extrapolatec retention time with a Vr window of2.1Vis longer than 10 years.

Impact ofthe trapped charge at the front O/N/O on the back-channel Vr.A VFT window of1.2 V provides a VBTshift of0.2V.

(a) Capacitorless IT-DRAM characteristics for NVM '0' and NVM 1' states. The similar sensing characteristics are achieved regardless ofthe data states ofNVM. (b) Data retention characteristics shows that the retention time ofNVM '(' shows longer than that ofNVM '1' because the charged electron retains the excess holes in the body.

VFT shift caused by charge trapping in the front O/N/O layer after cyclic capacitorless IT-DRAM operation. The soft-programming was found to be negligible.

Comparison of URAM for a multi-functional chip. In the separated double- gate URAM, the memory capacity can be doubled via simultaneous operation ofNVM and capacitorless IT-DRAM.

Schematic and SEM images ofthe FinFACT: (a) schematic ofaflip or flop state for the FinFACT and the tilted SEM image of(b) the initial straightened state, (c) the flipped or pull-up state, and (d) the flopped or pull-down state.

Fabrication flow: (a) fin patterning with SiN hard mask, (b) gate oxide and gate poly-Si deposition, gate separation byCMP, (c) gate patterning and S/D formation, (d) releasing sacrificial oxide at the NEMS area except the protected CMOS area, (e) re- oxidation and forming-gas annealing.

(a) List of the device dimensions, (b) cross-section TEM images of independently-controlled double-gate FinFET, and (c) cross-section TEM images of the FinFACT.

Structure of sense amplifier whit the independently controlled double-gate FinFET[8-9]. Usingthe independent gate operation, the current difference in the two pull- down paths is achieved by using a single ICDG FinFET in each path. The front gates ofNl and N2 are connected in the cross-coupled inverter configuration whereas BLB and BL are connected to the back gates.

ICDG FinFET CMOS characteristics for peripheral circuit application: (a) transfer characteristics and (b) output characteristics. The threshold voltage and drive current are modulated by the back-gate (G2) voltage.

Mechanical state ofNEMS transistor with the flip-flopped fin for memory application: (a) schematic, (b) symbol, and (c) SEM images. The fin remains straightened at the initial state. Bit 1' and bit '0' are distinguished when the fin is in contact with G1 and G2,respectively.

Drain current versus gate 1 voltage characteristics for transition from the pull- down state to the pull-up state. Gate2 is grounded during the measurement. Pull-out ofthe fin from the gate 2 electrode andpull-in to the gate1 electrode occur spontaneously.

(a) Drain current versus gate 1 characteristics at the pull-up (closed) and pull- down (open) states. Gate 1 and gate 2 serve as a drive gate and ground gate, respectively. At the pull-up state, a normal MOSFET curve is shown. At the pull-down state, a drain current level of off-state leakage is sustained regardless of gate 1 voltage, because the grounded gate2 fully depletes the fin. (b) Drain cu

Write voltage for initialization and bit change. At Wrm>60nm, the low adhesion force and high restoring force return the fin to the straighten state.

(a) Data retention characteristics and (b) data loss mechanism, especially at the wide fin. The strong adhesion force at WFm=30nm sustains a data state ofover 104 sec, but the mechanical restoring force gradually detaches the fin from the gate at Wrm=50nm, which results in reduction ofsensing window.

(a) Endurance characteristics and (b) SEM image of the failed device. At WFm=50nm, the operation over 103 cyclesis guaranteed. However, the mechanical fragility at WFm=30nm leads to fatigue-induced fin breakdown. This constraint can be relieved by optimizing device dimensions such as reduction ofair-gap.

Trade-off relationship between retention time and endurance for various fin widths. The compromised dimension is found to be Wrm=50nm.

The pull-up voltage required to switch the nanowire is calculated. The calculation expects that the operational voltage can be reduced according to the miniaturization ofthe device geometry.

Numerical analysis results ofthe fin deflection with transient programming Write speed of5 nsec is expected at Wrm=50nm.

A schematic of nano switch with encapsulating oil ambient. The used insulating liquid is highly refined oil free ofmoisture.