15 August 2017 in Scientific Reports
Neuro-biology inspired Spiking Neural Network (SNN) enables efficient learning and recognition tasks. To achieve a large scale network akin to biology, a power and area efficient electronic neuron is essential. Earlier, we had demonstrated an LIF neuron by a novel 4-terminal impact ionization based n+/p/n+ with an extended gate (gated-INPN) device by physics simulation. Excellent improvement in area and power compared to conventional analog circuit implementations was observed. In this paper, we propose and experimentally demonstrate a compact conventional 3-terminal partially depleted (PD) SOI- MOSFET (100 nm gate length) to replace the 4-terminal gated-INPN device. Impact ionization (II) induced floating body effect in SOI-MOSFET is used to capture LIF neuron behavior to demonstrate spiking frequency dependence on input. MHz operation enables attractive hardware acceleration compared to biology. Overall, conventional PD-SOI-CMOS technology enables very-large-scale-integration (VLSI) which is essential for biology scale (~1011 neuron based) large neural networks.
13 February 2018 in IEEE Electron Device Letters
Resistance random access memories (RRAM) or memristors with an analog change of conductance are widely explored as an artificial synapse, e.g., Pr0.7Ca0.3MnO3 (PCMO) RRAM-based synapses. In addition to synapses, scaled neurons are essential to enable a neuromorphic hardware. In this letter, we propose a PCMO RRAM for integrate and fire (IF) neuron. The analog conductance increase during SET process enables integration function. Upon exceeding a conductance threshold (i.e., fire) during a READ operation, a hard RESET is performed to reduce the conductance. The SET, READ, and RESET are performed in different phases of a clock to enable a PCMO for IF neuron. The availability of a non-volatile PCMO-based synapse makes PCMO for IF neuron attractive. Finally, PCMO-based neuron in spiking neural network yields software-equivalent classification accuracy as demonstrated on standard Fischer's Iris flower data set.
24 August 2012 in IEEE Electron Device Letters
We propose an epitaxial punchthrough diode for bipolar resistance RAM (RRAM) selector application. Epitaxial Si:C process is used to deposit n+/p/n+ layers which are fabricated into 300-nm-diameter vertical punchthrough diodes. High on-current density of >; 1 MA/cm2 and high on/off current ratio of >; 250 and >; 4700 (at opposite polarities) are observed. A switching speed of <; 10 ns is measured. On-voltage designability is demonstrated by tuning the p-region doping and length. The comparison of experimental IV with Sentaurus TCAD-simulated IV characteristics confirms the punchthrough mechanism. Comparison with other bipolar RRAM selector technologies highlights the overall advantages of punchthrough-based selector.
29 October 2012 in Applied Physics Letters
PMPrashanth Paramahans Manik
Prashanth Paramahans Manik
+3
We report low resistance Ohmic contacts on n-Ge using a thin ZnO interfacial layer (IL) capped with Ti. A post metallization anneal is used to create oxygen vacancies that dope ZnO heavily n-type ( ). Rectifying Ti/n-Ge contacts become Ohmic with 1000× higher reverse current density after insertion of IL. Specific resistivity of is demonstrated on epitaxial layers. Low resistance with ZnO IL can be attributed to (a) low barrier height from Fermi-level unpinning, (b) good conduction band alignment between ZnO and Ge, and (c) thin tunneling barrier due to the doping.
30 November 2006 in IEEE Transactions on Electron Devices
The three-dimensional (3D) electrostatics together with the modified Wentzel-Kramers-Brillouin tunneling model has been implemented to simulate the programming and retention characteristics of the metal nanocrystal (NC) memories. Good agreements with experimental data are first demonstrated to calibrate the transport parameters. In contrast to previous works, the 3D electrostatic effects investigated in this paper are proven very significant in the memory operations. Therefore, new design criteria of metal NC memories are investigated. Part I presents the physical model and the NC array design optimization. A sparse and large-size NC array, which is suitable for the one-dimensional narrow-channel memories, provides higher program/erase tunneling current density due to the field-enhancement effect and lower charging energy due to the large NC capacitance. On the other hand, to achieve a sufficient memory window, fast programming speed, and long retention time in the typical two-dimensional channel memories, a dense and large-size NC array is favorable while taking the tradeoff with the NC number density into account. Based on the same theoretical model, the authors continue in Part II to consider the design optimization when high-K dielectrics can be employed
19 April 2017 in IEEE Electron Device Letters
In the literature, various pulse-based programming schemes have been used to mimic typical spike time-dependent plasticity (STDP)-based learning rule observed in biological synapses. In this letter, we demonstrate the capability to generated arbitrary STDP behaviors by using analog programming waveforms inspired by neuronal action potential. First, we propose a simple algorithm to generate any arbitrary form of STDP. Second, we show the feasibility of a range of spike correlation time scales for STDP, e.g., biological ( ~100 ms) to accelerated (~20μs), based on W/Pr 0.7 Ca 0.3 MnO 3 /Pt based memristor. Third, we experimentally demonstrate several forms of STDP behaviors, where the preand post-neuronal waveforms are randomly spaced in time, akin to operational conditions. STDP shape corresponds well to waveforms. Thus, we show that artificial synapses can achieve the richness observed in biology as well as a range of STDP timescales for biologically compatible to accelerated neural network applications.