You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Any thoughts on this subject and where the technology is now would be appreciated. Could this be the best way of cooling chips and reduce the costs data centers incur in cooling the data center environment and maybe even recovering lost power? Thanks
I think that as chips cease to expand horizontally and we begin to see more layering, that simple radiative cooling will no longer suffice given the heat density. Peltier effect coolers and nano-scale heat pipes both have their pros and cons. Ultimately though, yes, I suspect moving power vias to the back will ultimately make room for front-side heat-transfer lanes.
The process described in the paper isn't exactly ready for mass production:
Clean sample holder; with acetone then dry it with N2. Fix it in the spinner. Spread aluminum foil in the spinner as shown in figure 10. Place glass slide with sample on the sample holder. Center it so that the center of the samplecoincides with the center of the holder. Press Vacuum ON by clicking HOLD. Check to make sure it is fixed properly and centered. Then take photoresist in a pipette and start covering the sample with it. Make sure to removethe bubbles first; cover the sample fully and as evenly as possible with the photoresist.Uniformly cover the area of the sample with drops of photoresist and wait for it cover thesample Figure 11. Fill any uncovered areas. Be careful sure not to touch the sample with the tip of the pipette. Load program BOYA4000 then click start centering. This rotates the sample at 4000RPM whichgives adequate thickness of photoresist on the sample.
Any thoughts on this subject and where the technology is now would be appreciated. Could this be the best way of cooling chips and reduce the costs data centers incur in cooling the data center environment and maybe even recovering lost power? Thanks
Given that thermocouples/thermoelectric coolers are very inefficient and power-hungry, using them to try and cool "hot chips" will increase overall power dissipation, not reduce it... :-(
Given that thermocouples/thermoelectric coolers are very inefficient and power-hungry, using them to try and cool "hot chips" will increase overall power dissipation, not reduce it... :-(
I was looking at thermocouples to actually convert the heat to electricity. I feel this technology is due for some serious research and become a major area of study. Much of the current technology in this area is aging and now the demand for solutions might push this technology forward. Especially with advances in AI/ML research should speed up dramatically.
I was looking at thermocouples to actually convert the heat to electricity. I feel this technology is due for some serious research and become a major area of study. Much of the current technology in this area is aging and now the demand for solutions might push this technology forward. Especially with advances in AI/ML research should speed up dramatically.
You need an order of magnitude improvement in the materials figure of merit (currently less than 2, after >200 years), you need a large temperature differential (not possible with thin film construction), and you need a hot side temperature greater than 100C.
None of these parameters are remotely compatible with current chip designs.
With the current state of the art materials, a Seebeck generator (many millimeters thick) attached to a chip running at 100C, sitting on a heat sink at 25C, would have less than 1% efficiency.
I was looking at thermocouples to actually convert the heat to electricity. I feel this technology is due for some serious research and become a major area of study. Much of the current technology in this area is aging and now the demand for solutions might push this technology forward. Especially with advances in AI/ML research should speed up dramatically.
