Improving the accuracy and robustness of RRAM-based in-memory computing against RRAM hardware noise and adversarial attacks

We present a novel deep neural network (DNN) training scheme and RRAM in-memory computing (IMC) hardware evaluation towards achieving high robustness to the RRAM device/array variations and adversarial input attacks. We present improved IMC inference accuracy results evaluated on state-of-the-art DNNs including ResNet-18, AlexNet, and VGG with binary, 2-bit, and 4-bit activation/weight precision for the CIFAR-10 dataset. These DNNs are evaluated with measured noise data obtained from three different RRAM-based IMC prototype chips. Across these various DNNs and IMC chip measurements, we show that our proposed hardware noise-aware DNN training consistently improves DNN inference accuracy for actual IMC hardware, up to 8% accuracy improvement for the CIFAR-10 dataset. We also analyze the impact of our proposed noise injection scheme on the adversarial robustness of ResNet-18 DNNs with 1-bit, 2-bit, and 4-bit activation/weight precision. Our results show up to 6% improvement in the robustness to black-box adversarial input attacks.

Optimization of triple-alternating-gas (TAG) injection technique for enhanced oil recovery in tight oil reservoirs

After single-gas (SG) injection operations in tight oil reservoirs, a significant amount of oil is still unrecovered. To increase productivity, several sequencing gas injection techniques have been utilized. Given the scarcity of research on multiple-gas alternating injection schemes, this study propose an optimized triple-alternating-gas (TAG) injection for improved oil recovery. The performance of the TAG process was demonstrated through numerical simulations and comparative analysis. First, a reservoir compositional model is developed to establish the properties and composition of the tight oil reservoir; then, a suitable combination for the SG, double alternating gas (DAG), and TAG was selected via a comparative simulation process. Second, the TAG process was optimized and the best case parameters were derived. Finally, based on the oil recovery factors and sweep efficiencies, a comparative simulation for SG, DAG, and TAG was performed and the mechanisms explained. The following findings were made: (1) The DAG and TAG provided a higher recovery factor than the SG injection and based on recovery factor and economic advantages, CO2 + CH4 + H2S was the best choice for the TAG process. (2) The results of the sensitivity analysis showed that the critical optimization factors for a TAG injection scheme are the injection and the production pressures. (3) After optimization, the recovery factor and sweep efficiency of the TAG injection scheme were the best. This study promotes the understanding of multiple-gas injection enhanced oil recovery (EOR) and serves as a guide to field design of gas EOR techniques.

A Tracer Streamline Practice for Re-Evaluation Waterflood Pattern to Introduce a Cyclic Water Injection Scheme

Waterflood introduces in the oil field a couple of years ago. Several waterflood schemes have been implemented in the fields to get the best incremental oil, such as peripheral injection, pattern waterflood, and etcetera. Many waterflood schemes are not working properly to boost the oil recovery due to unpredicted and unexpected water tide array. Then, the tracer practice started to be used for getting a better picture of the transmissibility reservoir as well as the direction of water pathway. This practice honors the parameters, such pressure, water cut, GOR, and rates. The streamline modeling is used to map the tracer, and it concludes that the selection of location of the injector should be based on the highest oil recovery achieved. Subsequently, the cyclic water injection method is one alternative. Apparently, this approach yields a quantify incremental recovery. This research utilizes the pressure different approach to figure out the route of water in the formation. The inter-well tracer technique in this modeling study is a tool to review communication between injectors and producers in the existing pattern. Many scenario should be tried to find the best options for the new pattern opportunities. In parallel, a innovative scheme of waterflood technique should be implemented too for escalating oil recovery. The stream pathway observes a new potential of the waterflood scheme. It is called "cyclic injection" scheme. The novelty of this approach is the ability to solve the poor sweep efficiency due to improper pathway of water influx in the oil bearing".

Design study of CEPC lower emittance booster

The CEPC booster needs to provide electron and positron beams to the collider at different energy with required injection efficiency. At Higgs energy, only the on-axis injection from booster to collider can be fulfilled in CDR. With a consideration of keeping the off-axis injection scheme for safety and reliability, a new booster design based on TME lattice is considered to reduce the emittance by three times after CDR. The new booster design has reached an emittance of 1.3 nm at 120 GeV and the DA without errors is even better than CDR. The geometry of new booster is designed carefully in order to share the same tunnel with collider. The design status of CEPC new booster including parameters, optics, dynamic aperture and geometry is discussed in this paper.