Energy-Efficient Multiobjective Thermal Control for Liquid-Cooled 3-D Stacked Architectures
read more
Citations
Wireless Sensor Network Optimization: Multi-Objective Paradigm.
3D-ICE: A Compact Thermal Model for Early-Stage Design of Liquid-Cooled ICs
A novel thermal management scheme for 3D-IC chips with multi-cores and high power density
Multi-objective optimization in sensor networks
References
Fuzzy identification of systems and its applications to modeling and control
High-performance heat sinking for VLSI
Stability analysis and design of fuzzy control systems
Temperature-aware microarchitecture
Related Papers (5)
Frequently Asked Questions (19)
Q2. What is the effect of changing the flow rate on the thermal development of the microchannel?
When the flow rate is increased, the fluid gains velocity and spends less time in the microchannel, thus causing the fluid to have a lower temperature at the outlet port in comparison to using a smaller flow rate value.
Q3. What is the effect of interlayer cooling on the thermal gradient between the chip and the liquid?
In fact, since the liquid has a lower temperature when it is injected in the microchannel, the thermal gradient between the chip and the liquid is larger and more heat can be absorbed by the liquid.
Q4. What is the effect of interlayer cooling on the thermal gradient of the stack?
Although interlayer liquid cooling mitigates the thermal hot spots, it increases both the intralayer and the maximum thermal gradient of the stack.
Q5. How much heat is removed at interconnect pitches?
At a chip size of 1 cm2 and maximal difference between junction and liquid cooling temperatures of 60 K, the heatremoval performance is more than 200 W/cm2 at interconnect pitches larger than 50 μm.
Q6. How does the author propose to reduce the thermal simulation time in HotSpot?
To reduce the potentially long thermal simulation time in HotSpot for large MPSoCs, recentwork proposes a thermal emulation framework using fieldprogrammable gate arrays [2].
Q7. How does the thermal reduction of the cores affect the flow rate?
In fact, this thermal reduction reaches values up to 40 °C when the flow rate is changed from the minimum to maximum values because the flow thermal development depends on the flow rate.
Q8. What is the effect of FALB on the thermal gradient of a 3-tier MPSoC?
This increase in thermal gradient is correlated with the cooling technique, which includes single heat sink on the top layer, hence creating a huge thermal gap between tiers of different relative distances with respect to the heat sink.
Q9. How many times do the authors sample the utilization percentage for each hardware thread?
The authors sample the utilization percentage for each hardware thread at every second using mpstat, and record half an hour long traces for each benchmark.
Q10. What is the effect of changing the flow rate on the thermal development of the cores?
Based on these results, it is highly beneficial to increase the flow rate when the cores located away from fluid inlet have high temperatures, enabling temperature reduction regardless of the activity of the other cores.
Q11. What is the effect of FALB on the thermal gradients of the systems?
By applying FALB, the thermal gradients of the systems are significantly reduced, implying a better thermal balance (Figs. 13 and 14).
Q12. How do the authors compute the junction temperature of 3-D MPSoCs with liquid cooling?
The authors compute the local junction temperature of 3-D MPSoCs with liquid cooling using the resistive network of Fig. 2 for each layer, where heat is dissipated to both opposing vertical directions (up and down) from the heat sources.
Q13. What is the performance degradation of the average workload under a set of policies?
The performance degradation of the average workload under a set of policies is shown inThe authors have shown in their previous results that in the 4-tier 3- D MPSoC case, the maximum temperature is kept below the thermal threshold (85 °C).
Q14. How can the back-side liquid cold plates handle up to 400 W/cm2?
Recent work shows that back-side liquid cold plates, i.e., staggered microchannel and distributed return jet plates, can handle up to 400 W/cm2 in single-chip applications [6].
Q15. What is the proposed technique for 3-D MPSoCs with liquid cooling?
Based on the previous analysis, the authors propose the flow-aware 3-D load balancing (FALB) technique, a novel scheduling technique for 3-D MPSoCs with liquid cooling, where the tasks requiring higher throughput are allocated on the processing elements based on their distance to the fluid inlet port.
Q16. How much energy is required to inject the fluid to all stacks?
The maximum energy required to inject the fluid to all stacks is approximately 180 W, which is a significant overhead to the whole system as this value is similar to the energy consumed by a 4-tier 3-D chip.
Q17. What is the effect of VF scaling on the cores near the outlet port?
This behavior is related to the impact of VF scaling on these cores: when the VF is scaled down to reduce the temperature, DVFS has a higher impact on the cores closer to the inlet port than those further from the inlet port.
Q18. How do the cores near the outlet port perform VF scaling?
as shown in Fig. 8, the cores located at 0.7 mm from the inlet port have fewer VF changes in comparison to the cores at 0.5 mm.
Q19. What is the tier of the 3-D test chip?
The 3-D MPSoC the authors use for the analysis consists of two tiers, where each tier contains four main hot spot sources modeling high performance processors.