Search

Recently, artificial intelligence has made rapid progresses. However, existing AI systems still encounter difficulties even for somethings that humans can easily do. The ultimate way to solve these problems is to develop artificial general intelligence (AGI). Brain inspired computing (BIC) systems are one of the most promising technologies to integrate computer science and neuroscience to facilitate the development of AGI. In this talk, three issues will be discussed: (1) why do we need BIC system? (2) the current status and latest progress in BIC chips, software, and systems; (3) how to develop BIC systems to support AGI with limited understanding of the brain mechanisms. A hybrid and scalable exploration platform of AGI is demonstrated by an unmanned bicycle control system. The main challenges, possible solutions and strategies to develop BIC systems to stimulate AGI development will be addressed.

Brain inspired computing(BIC) system is one of the most promising technologies for the development of artificial general intelligence (AGI). In this talk, the recent development of neuromorphic devices, chips and systems are reviewed. The three key issues will be discussed: (1) why do we really need brain inspired computing system? (2) What is current status for BIC? (3) how to build BIC systems without a complete understanding of the brain mechanisms? The possible solutions and strategy to develop BIC system will also be addressed.

Although a variety of solutions for neuromorphic systems based on different hardware technology and software programming schemes, there has yet to be a common accepted one. Based on some recent findings in brain science, we propose a new design rule for developing a brain inspired computing system. We design and fabricate a neuromorphic chip, named ‘Tianji’ chip. A multi-chip architecture-based PCB board has been demonstrated. The detailed hardware implementation and software programming scheme are presented in this paper.

Brain inspired computing system is very important for building up real intelligent systems. In this talk, the current status and the recent progress of brain inspired computing research are introduced. The key challenges of this research including fundamental theory, neuromorphic chip design and fabrication, and software environment are analyzed. The possible solutions, development strategy and roadmap, and future trends of this research are discussed.

A model for encoding and the retrieve of the sequential working memory is proposed by using bidirectional inhibition-connected neural networks with winnerless competition. It is found that the retrieve accuracy is dependent on the encoding time the the properties of the neural inhibition weights. The simulation results shows the effectiveness of our proposed model.

The development of conventional computer is limited by the separated CPU and memory, as well as the desperate Moore's law. It's a trend to develop a brain-like computer architecture in order to solve above problems, where a new memory is highly needed. Artificial cognitive memory (ACM) based on neuromorphic circuits, integrates computation tightly with storage like the brain and is the key component to support brain-like computation. The power issue of ACM is critical, especially for the application of portable devices. In this paper, we systematically analyze the key factors of the energy consumption for building an ACM chip, such as element model, silicon technology, synaptic material, array selecting, weight modification method, routing strategy and the integration approaches. These results provide an in-depth insight for creating an ultra-low power brain-like computing platform.

Revised algorithm for online learning with kernels (OLK) in classification and regression is proposed in a reproducing kernel hilbert space (RKHS). Compared with the original OLK, the revised algorithm allows that the new data points arrive either one by one or two by two.

Phase-change memory (PCM) represents one of the best candidates for a “universal memory”. However, its slow SET speed, high RESET power, and high resistance drift present key challenges towards this ambition. Here, grain-engineered Ge 2 Sb 2 Te 5 is exploited to control the crystallization kinetics, and electrical properties of PCM. We report 120 % higher SET speeds with respect to conventional scaling. Good stability (140 °C), 30 % RESET power reduction, and 2X lower resistance drift were also achieved. A 4-state/2-bit multilevel cell was further demonstrated. This provides a route to making high-density PCM devices.

Al/Ta2O5/Pt RRAM cell was successfully demonstrated with stable bipolar switching. Band alignment for the structure of Al/Ta2O5/Pt was obtained using high-resolution X-ray photoelectron spectroscopy to investigate resistance switching behavior. Hole barrier heights of Ta2O5 on metals Pt and Al were extracted using core level and valence band spectra. The energy band alignments of Pt/TaOx and Al/TaOx were thus determined and correlated to conduction mechanisms governing SET and RESET processes of actual bipolar Pt/TaOx/Al devices.

The conventional view of axon as a transmission cable has been challenged by continuous progress in neuroscience discoveries, which indicate the rich functional and computa-tional repertoire of the axon. Recent experimental findings of slow integration induced persistent firing in distal axons of interneurons have shown that the slow integration from tens of seconds to minutes in distal axon leads to persistent firing of action potentials lasting for similar duration, suggesting that the axon performs its own integration functions. In this paper, we present an artificial neuron model including both somatic and axonal computation units, which reproduces the neural behav-ior of persistent firing. Complementary to the classic somatic computational unit which evokes action potentials by integrating dendritic inputs in a short timescale, the axon integrates the soma evoked spikes in a longer timescale of tens of second to minutes. Consequently the persistent firing behavior of the axon is determined through toggling the axon dynamics between passive conduction mode and persistent firing mode based on the integrated axonal potential. We present and discuss in this work the mathematical and neuromorphic models of the artificial neuron, as well as their simulation results. The artificial neuron proposed, being computationally efficient yet bio-plausible, would be useful to construct and simulate the large scale models of animal or human cortex, which provides a neuromorphic platform for further investigation of the possible functions of persistent firing and their roles in animal and human brain, especially their correlations with working memory.

The thermal conductivity of nitrogen-doped Ge 2 Sb 2 Te 5 (N-GST) was analyzed using Frequency Domain Thermoreflectance (FDTR). The thermal conductivity of amorphous N-GST (∼0.15 W/m-K) was not found to change significantly as the nitrogen concentration was raised from 0 at% to ∼6 at%, possibly due to the huge amount of phonon scattering in the disordered films. The thermal conductivity of crystalline N-GST films was found to increase initially with increasing N content, but then to decrease upon further N addition. X-ray diffraction spectra of N-GST films show increasing defect density that correlates with the decrease in thermal conductivity of the crystalline films at higher nitrogen content.

Films of crystalline and amorphous Fe 12 (Ge 2 Sb 2 Te 5 ) 88 were prepared using laser synthesis from a stoichiometric target. The crystalline film, prepared at 300 °C was found to be in the hexagonal close packed (HCP) structure while the film prepared at room temperature was found to be amorphous. The crystalline film showed clear magnetic hysteresis, while the amorphous and the 220 °C prepared film exhibited weaker magnetic properties. The crystallization temperature of the amorphous Fe 12 (Ge 2 Sb 2 Te 5 ) 88 as extracted from resistance temperature measurements was found to be 168 °C, comparable to 178 °C for pure Ge 2 Sb 2 Te 5 . Lastly, the reflectivity spectrum showed the typical reflectivity contrast between crystalline or amorphous Fe 12 (Ge 2 Sb 2 Te 5 ) 88 films — as expected from phase change films.

In this paper, we report a novel Cr/TaO x /Al (top to bottom) Resistive Random Access Memory (RRAM) which works as a bipolar switching device. The RRAM devices have demonstrated excellent memory performance including small magnitude of switching voltages (about 2 V), a tight distribution of V set and V reset , low switching current, large off/on resistance ratio R of up to 107, and good retention characteristics (more than 105 s) at high temperature (120°C). Resistance of both states shows little degradation, and retention characteristics can be extrapolated to 10 years. The relationship between active area and resistance at low resistance state is studied and the switching appears to be a local phenomenon which is likely to be of the filament type. Forming gas anneal has an additional positive effect on the device performance parameters, as it leads to smaller magnitude of switching voltages and better uniformity of the resistance in the high resistance state.

We report the first demonstration of phase change memory cells utilizing nickel monosilicide (NiSi) or platinum monosilicide (PtSi) as the bottom electrode or heater material. This work enables the integration of PCRAM directly on the silicided drain regions of MOSFETs, allowing compact integration with reduced process complexity and cost. Low reset current of 0.8 mA was achieved for contact dimensions of ~ 1 μm. Electrical and simulation results demonstrate the feasibility of employing silicides as a bottom electrode/heater in a PCRAM.

A multi-layer lateral PCM with N-doped Sb 70 Te 30 was proposed and demonstrated. Both current sweep and pulse mode dynamic resistance test show that multi states exist in the device, which can be used for multi-level data storage. Simulation shows the working mechanism of multi-level and confirms the experiment results. More intermediate states can be realized by increasing the cycles of N-doped Sb 70 Te 30 and ZnS-SiO 2 deposited and by using different film thickness, which will be a promising solution to increase the data storage capacity for PCM largely.

We report the first demonstration of a strained In 0.53 Ga 0.47 As n-MOSFET incorporated with a silicon nitride (SiN) liner stressor. High intrinsic tensile stress of 1.5 GPa in a 50 nm thick SiN was used for introducing lateral tensile strain in the In 0.53 Ga 0.47 As channel. SiN liner stressor was shown to provide significant drive current enhancement for the first time. In addition, an advanced gate dielectric technology (with SiH 4 + NH 3 treatment) was used for achieving low interface state densities for HfAlO gate dielectric formed on In 0.53 Ga 0.47 As. Strained In 0.53 Ga 0.47 As n-MOSFETs with good device performance are reported.

Femtosecond laser is a unique processing tool to make micro-/nanostructures on heat sensitive bio-polymer film & glass substrate, to create 3D structures inside glass, to achieve 20-nm nanostructuring and large-area THz, micro-mold and micro-device patterning.

Phase change RAM (PCRAM) is one of the best candidates for the next-generation nonvolatile memory. Recently lateral PCRAM using a thin phase change bridge was proposed as a promising approach to achieve high density due to simpler fabrication process and lower RESET current. This paper proposes a new lateral PCRAM structure - edge contact lateral structure, together with a Sb 7 Te 3 -GeTe super-lattice-like (SLL) phase change medium to reduce the contact area, improve thermal confinement and hence reduce current. Its RESET current of 1.23 mA is less than that of normal lateral PCRAM with SLL (1.5 mA). It also shows good stability and resistance ratio after 105 overwriting cycles. Testing results are consistent with the simulation results.

Development of modern materials, including nanoclusters, cluster assembled materials and metamaterials is among the actual challenges for the development of future nanotechnologies. Here we discuss the peculiarities of far-field and near-field light scattering by plasmonic nanoparticles, and possible applications of weakly dissipating materials. Over the last few years many peculiarities of light scattering have been found for nanoparticles in the regime of plasmon resonances. Optical excitation of localized plasmons is accompanied by inverse process - transformation of localized resonant plasmons into scattered light. When radiative damping prevails over the dissipative damping, the effects of anomalous light scattering result in sharp giant optical resonances and complicated near-field structure of the Poynting vector field, see e.g. [1-4]. Here we present peculiarities of far-field and near-field light scattering by plasmonic nanoparticles with weak dissipation and anisotropy.

We report a novel surface passivation technology employing a silane-ammonia gas mixture to realize very high quality high-k gate dielectric on GaAs. This technology eliminates the poor quality native oxide while forming an ultrathin silicon oxynitride (SiOxNy) interfacial passivation layer between the high-k dielectric and the GaAs surface. Interface state density Dit of about 1 times 1011 eV-1 cm-2 was achieved, which is the lowest reported value for a high-k dielectric formed on GaAs by CVD, ALD, or PVD techniques. This enables the formation of high quality gate stack on GaAs for high performance CMOS applications. We also realized the smallest reported (160 nm gate length) inversion-type enhancement-mode surface channel GaAs MOSFET. The surface-channel GaAs MOSFETs in this work has demonstrated one of the highest peak electron mobility of ~2100 cm2/Vmiddots. The lowest reported subthreshold swing (~100 mV/decade) for surface-channel GaAs MOSFETs was also achieved for devices with longer gate length. Extensive bias-temperature instability (BTI) characterization was performed to evaluate the reliability of the gate stack.

In this paper, phase change random access memory (PCRAM) devices with two different geometric configurations were studied. The geometry effect on device functionalities in terms of thermal and electrical properties has been analyzed by simulation and experiments. Based on 0.35mum technology, 128 bits PCRAM array integrating with CMOS as the selection device were fabricated. The two types of PCRAM cells were implemented and characterized. The simulation results showed that both the bottom electrode and phase change layer geometry and the size of the interface between the phase change material and the bottom electrode were the determinant factors to the temperature profile and heat distribution. Both simulation and experimental results show that PCRAM device with type II structure required lower current to start phase change from crystalline state to amorphous state. However, it required a larger current to be fully RESET. A model was proposed in this paper to discuss this phenomenon

In this work, phase change random access memory (PCRAM) was studied theoretically and experimentally. Phase change materials were deposited and their physical parameters were measured. A simulation and design software for PCRAM was developed based on multidisciplinary theories including electrodynamics, thermal conduction, crystallization kinetics and numerical computations. By introducing physical models of PCRAM elements, a general macromodel of the phase change random access memory (PCRAM) elements for HSPICE-based computer simulator is proposed. PCRAM array were designed, fabricated, and tested by using a self built tester. Also, near field optical scan microscope incorporated with fs laser was used to fabricate nano scale PCRAM cells

Understanding of the phase change in nano-scale, or so-called nano-phase change, and its related issues are important for its applications on both optical recording and phase change random access memory (PCRAM). Nano-phase change can be classified into thickness-dependent and structure dependent types. For PCRAM device performance, the film thickness-dependent thermal profile and materials' properties are the two important factors. In this work, the thickness dependence of nano-phase change for chalcogenide materials was studied by both simulation and experiments. A thermal model was built up to simulate the film thickness-dependent thermal profiles and its corresponding effects of line type PCRAM. The simulation results showed that the temperature profile, heating rate and cooling rate are strongly dependent on the thickness of phase change materials. Experiments had been conducted to investigate the thickness dependence of crystallization temperature for both nucleation-dominated and growth-dominated phase change material. It was found that the crystallization temperatures are function of the thickness for the both materials. Higher crystallization temperatures were obtained for thinner phase change films. The influence of above factors on the performance of PCRAM devices is discussed.

A thermal analysis of chalcogenide random access memory (C-RAM) cell, a non-rotational nonvolatile phase change memory cell, was conducted. A three-dimensional finite-element method was used in the simulation of C-RAM cell. The thermal effect generated by applying an electric pulse was calculated and analyzed. The dependence of its thermal performances due to electrical and geometrical variations was also investigated. Temperature performances including temperature profiles and history, the heating and cooling rates, and heat flow characteristics were obtained and analyzed.

We present a two terminal HSpice model for chalcogenide based phase change memory (CRAM) element. By including physical models of CRAM programming, this model can simulate not only the resistance change by different electrical pulses, but also temperature profile and crystalline fraction during the operation. Furthermore, it was successfully integrated with standard W/R circuit in memory technology. Output of sense amplifier vs writing current amplitude figure corresponded well with the typical R-I curve of CRAM elements.

High power blue light generated by quasi phase matched (QPM) second harmonic generation (SHG) techniques is very attractive for high density optical storage. KTiOPO/sub 4/ (abbreviated as KTP) is one of the most important materials used in QPM SHG. Since 1987 many kinds of waveguides and devices have been demonstrated in KTP substrates. However there still remains some technique problems, for example, the control of the duty cycle of the segmented waveguides as well as cracking of the waveguides. On the other hand the photolithography process used for fabrication of KTP segmented waveguides has not been reported. In this paper we report on the experimental results of segmented optical waveguides fabricated in KTP substrates.