Narrow Results

This paper uses the finite volume lattice Boltzmann method (FVLBM) to simulate the transient heat conduction from macro- to nano-scales corresponding to Kn = 0.01–10. This model is used for two dimensional (2D) transient hotspot modeling. The results of the diffusive regime are compared with those of the Fourier law as a model of continuum mechanics and an excellent agreement is found in this regime. After the validation of model for the case of Kn = 0.01, it has been used for high-Knudsen number simulations and a test case with Kn = 10 is studied. By increasing the Knudsen number from 0.01 up to 10, the transition from totally diffusive to totally ballistic behavior has been discussed and the wave-feature of heat transport through the solid material has been investigated.

Miniaturization and the continued scaling of CMOS technology leads to the high-power dissipation and ever-increasing power densities. One of the major challenges for the designer at all design levels is the temperature management, particularly the local hot spots along with power dissipation. In this work, the controller circuits which are implemented as Finite State Machines (FSMs) are considered for their thermal-aware and power-aware realization. Using Genetic Algorithm (GA), both encoding and bipartitioning of the FSM circuit are implemented to get two subFSMs such that at a particular instant of time, one subFSM is active at a time, whereas the other one is power-gated. Again, thermal-aware realization (in terms of power-density) of this power-gated FSM is done. Therefore, the work concerns with the thermal-aware encoding and partitioning of FSM for its power-gated realization. Average temperature saving obtained in this approach for a set of benchmark circuits over previous works is more than 16%. After getting the final partitioned circuit which is optimized in terms of Area and power-density, thermal analysis of the sunFSMs is performed to get the absolute temperature. As thermal-aware design may increase the area, a suitable area-temperature trade-off is also presented in this paper.

In the nano-scale era, chip temperature has gained a lot of importance due to various reasons. Localized high temperatures in certain regions, commonly known as hotspots, directly affect the performance and reliability of the chip. The proposed work concentrates on temperature-driven floorplanning of chips because incorporating cooling techniques may lead to increase in cost. The proposed approach attempts to evenly distribute power dense components across the chip area in order to suppress hotspots and at the same time generate a layout with aspect ratio nearly 1, as is required in the present day floorplanning scenario. The positions of the blocks in the layout are generated using Boolean Satisfiability (SAT). Apart from peak temperature, wirelength being an important cost factor is also taken into consideration. The proposed technique is successful in decreasing the wirelength and peak temperature for large GSRC, MCNC and IBM HB$+$ benchmarks.

With an increase in microelectronic system density, using conventional technologies for cooling continues to become more challenging and is often a limiter of performance and efficiency. One key challenge is to address high background heat fluxes generated across entire chips and packages while mitigating localized hotspots with even higher heat fluxes. In this chapter, cooling of integrated circuits with non-uniform power maps using non-uniform micropin-fin heat sinks is investigated. Two heterogeneous micropinfin samples were fabricated, and single-phase experiments with deionized water were performed to investigate the effectiveness of local micropinfin clustering for hotspot cooling. Large background heat fluxes (250 W/cm²) were cooled in conjunction with even higher hotspot heat fluxes (500 W/cm²), with the lowest achieved average junction-to-inlet thermal resistance (R_th) of 0.092^°C/W. Subsequently, to demonstrate the benefit of microfluidic cooling on active silicon, a micropinfin heat sink was etched into the bulk of an Altera Stratix-V field-programmable gate array (FPGA) built using a 28 nm CMOS process. With a benchmark pulse compression algorithm running on the FPGA, thermal and electrical measurements were made. Deionized water was used as a coolant, with flow rates ranging from 0.15 to 3.0 mL/s and inlet temperatures ranging from 21^°C to 50^°C. An average junction-to-inlet thermal resistance of 0.07^°C/W or a heat transfer coefficient of 1.59 × 10⁴ W/m² °C was achieved.

Protein–nucleic acid interactions are inevitable in maintaining the homeostasis of cells. It is important to have a quantitative understanding of these interactions, generally described in terms of the dissociation constant or free energy change of protein–DNA and protein–RNA complexes. These interactions are impaired in the presence of mutations in nucleic acids or their interacting proteins, leading to numerous diseases. Hence, it is important to understand the binding affinity change upon mutation in protein–nucleic acid complexes. Different experimental techniques are available to study the binding affinities, although they are accurate, it is time and labor-intensive. On the other hand, computational techniques are emerging with numerous databases and computational tools to study protein–DNA complexes. In this chapter, we discuss various databases for the binding affinity of complexes and change upon mutation and the tools available to extract different structural and interaction features from the complexes. Further, we provide details on prediction methods reported for predicting the change in binding affinity upon mutation, along with hotspot residue prediction in protein–DNA complexes.

With the increase of urbanization in cities, there is also the rise of crime incidence. It is important for the police force to stay ahead in proactive policing. In this paper, the researcher presents a simple approach to predict crime in a geographic space using grid thematic mapping and neural networks. The study particularly focuses on the possibility of using historical crime data in predicting future crime incidents. The dataset is divided into monthly, weekly, and daily data called as a snapshot for training the model. The study area, Cebu City, is divided into a square grid of various sizes and crime incidents, from each cell in the grid of each snapshot, will then be recorded. This data is then fed into the neural network to predict possible areas where crime would happen for the next time interval. Initial results have shown that the model is accurate enough to predict crime areas when the data snapshot is divided monthly and weekly and grid size is set to a large size of about 1000 by 1000 meters and 750 by 750 meters. The best model was able to yield an F1 score of 0.95. Although the created model is simple, this can be a stepping stone to further crime modelling and prediction studies. This tool can be used to aid decisions and policy-making in the deployment of police resources throughout the city.