Ferroelectric FET analog synapse for acceleration of deep neural network training

Abstract

In this paper we propose and evaluate the performance of a 3D-embedded neuromorphic computation block based on indium gallium zinc oxide ( -IGZO) based nanosheet transistor and bi-layer resistive memory devices. We have fabricated bi-layer resistive random-access memory (RRAM) devices with Ta2O5 and Al2O3 layers. The device has been characterized and modeled. The compact models of RRAM and -IGZO based embedded nanosheet structures have been used to evaluate the system level performance of 8 vertically stacked -IGZO based nanosheet layers with RRAM for neuromorphic applications. The model considers the design space with uniform bit line (BL), select line (SL) and word line (WL) resistance. Finally, we have simulated the weighted sum operation with our proposed 8-layer stacked nanosheet based embedded memory and evaluated the performance for VGG-16 convolutional neural network (CNN) for Fashion-MNIST and CIFAR-10 data recognition, which yielded 92% and 75% accuracy respectively with drop out layers amid device variation.

Graphene–ferroelectric transistors as complementary synapses for supervised learning in spiking neural network

“…They are at the same levels with those recently reported in other kinds of ferroelectric synapses. 25 A conventional convolution neural network (CNN) is then set up where both the convolutional kernels and the connections in the fully connected layers are implemented with the GrFeFET synapses. By taking these nonideal parameters into account, the simulation yields a recognition rate of 94%, when implementing MNIST tasks (details are provided in Section "GrFeFET synapse in CNN for MNIST recognition" of Supplementary Information).…”

Phase field modeling of domain dynamics and polarization accumulation in ferroelectric HZO

“…Ferroelectric (FE) materials, particularly Zr doped HfO 2 (Hf 1-x Zr x O 2 :HZO 1 ) have drawn significant research interest in recent times due to CMOS process compatibility 2 , thickness scalability 3,4 as well as many promising attributes of ferroelectric field effect transistors 2,5 (FEFETs) for low-power logic [5][6][7][8] and non-volatile memories 9,10 applications. In addition, FEFETs can provide multiple non-volatile resistive states that harness the multi-domain FE characteristics, leading to the possibilities for multi-bit synapses 11,12 in a neuromorphic hardware 13 . Further, newly reported accumulative polarization (P)-switching process 14 in ultra-thin FE leads to many appealing opportunities for novel applications like correlation detection 15 and other non-Boolean computing paradigms 16 .…”

Memristive and CMOS Devices for Neuromorphic Computing

“…This three-terminal device has recently been operated into memory arrays with 28 nm CMOS technology [114] and exhibits a strong potential for the development of 3D structures [115]. Also, it has been operated to replicate synapse [116] and neuron [117,118] functions, which, combined with 3D integration opportunity, makes it a strong candidate for neuromorphic computing applications. Figure 12b illustrates the device structure of the ECRAM consisting of an MOS transistor where a solid-state electrolyte based on inorganic materials, such as lithium phosphorous oxynitride (LiPON) [108,119], or organic materials, such as poly (3, 4-ethylenedioxythiophene):polystyrene sulfonate (PEDOT:PSS) [120], is used as the gate dielectric.…”