Test Processor With New Thermal Transistors Cools Chip Without Moving Parts
Solving thermal density with new transistor architecture.
The latest processors are becoming hotter with almost every generation, but a study from the University of California L.A. shows that heat-channeling thermal transistors might be the solution (via IEEE). Although these thermal transistors are just in the experimentation stage, they offer a highly attractive way to remove heat from processors that may pique the interest of companies like AMD and Intel.
Today's modern processors, particularly the high-end ones, have an acute issue when it comes to heat. Processors are getting smaller, but power consumption isn't going down nearly as much, and since power becomes heat, that means more heat is being packed into a smaller area. This heat is often concentrated in a specific part of the processor (a hot spot), and even if the average temperature of a CPU is fine, the hot spot temperature may hold it back from performing as well as it could otherwise.
These new thermal transistors essentially channel that heat throughout the rest of the processor via the use of an electrical field, spreading out the heat evenly. The design innovation that made this possible was a one-molecule-thin layer of molecules that become thermally conductive when charged with electricity. Thermal transistors could move the heat from a hot spot (often in the cores) to a cooler part of the chip. Compared to normal cooling methods, the experimental transistors were 13 times better.
The advent of heat density issues can be traced back to Dennard Scaling. Dennard Scaling held that smaller transistors were more efficient, which meant heat density would never go up. However, Dennard Scaling stopped holding up in the mid-2000s, around the time the industry hit the 65nm process node. Ever since, the ratio of power/heat to area has been increasing gradually.
Additionally, as processor designers enhance their frequency-boosting algorithms to extract more performance, it's become more evident how hard it is to beat the heat. If thermal transistors make it out of the lab and into consumer devices, it might at least stave off the heat density problem, if not solve it outright. Otherwise, more exotic versions of traditional cooling methods might be necessary, like immersion cooling.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Matthew Connatser is a freelancing writer for Tom's Hardware US. He writes articles about CPUs, GPUs, SSDs, and computers in general.
Jobs eviscerated at Chinese Arm chip design firm in wake of restrictions from TSMC — lack of access to 7nm node could cause 150 employees to be laid off
Chinese chip firms say a new round of US sanctions won’t stop China’s chip industry — Chinese government responds with its own export restrictions anyways
-
chaz_music About 20 years ago, I talked with Nextreme (now Laird) about using their Peltier devices being used internally in processors and LED chips. Since the Pelitier material is semiconductor, someone should have figured out by now how to embed Peltier junctions where they are needed and create a thermal cooling channel on-die. Unless they did and the idea is not out in the wild yet?Reply -
Diogene7 Well really incremental improvements to silicon transistors is close to reaching its limit, and therefore it would be better to significantly invest in next generation computing technologies like spintronics related technologies (MRAM, Intel MESO concept,…) as it seems that could improve power efficiency from 5x to 30x…Reply
I don’t understand that at this stage DARPA and the US government (US CHIPS Act) still don’t yet make it a top priority for the US to regain its leadership… -
ekio 13 times cooler... what does it mean even ?Reply
The processor is 10 degrees instead of 130 without cooling system, it's 5 degrees compared to 65 with a classic cooler, it's 1/13th above room temperature, it's no cooler - room temp + 1(/13* room temp). etc ???
Explain properly what it means!
You guys put some blurry figures often and nothing is making any sense nor explained.... -
Ravestein NL Maybe it's possible to recycle the processor's power by using the heat to electricity way like those ventilators they make for wood stoves! And so lowering the power the CPU uses!Reply -
TJ Hooker
The 13x value is the difference in thermal conductance between the thermal transistors being in an 'on' vs 'off' state. Not the difference compared to conventional cooling methods, as incorrectly stated in this article.ekio said:13 times cooler... what does it mean even ?
The processor is 10 degrees instead of 130 without cooling system, it's 5 degrees compared to 65 with a classic cooler, it's 1/13th above room temperature, it's no cooler - room temp + 1(/13* room temp). etc ???
Explain properly what it means!
You guys put some blurry figures often and nothing is making any sense nor explained....
I'm not sure why that's an important metric though. At least in the context of cooling computer chips, simply having high thermal conductivity seems like key property, rather than the ability to change conductivity. The IEEE article also highlights the frequency at which the thermal conductivity can be switched, which I also don't understand the significance of. -
chaz_music Ravestein NL said:Maybe it's possible to recycle the processor's power by using the heat to electricity way like those ventilators they make for wood stoves! And so lowering the power the CPU uses!
I like that style! Your idea would be novel - except it causes heartburn. You are talking about embedding a thermocouple or other form of Peltier device inline with the CPU heat flow to extract energy. That comes at a cost.
Any energy conversion requires the same idea as a "voltage drop" and "current flowing through it". Heat is the same.
Let's start with thinking about motors. In motors the output mechanical power is P = torque x speed, which requires the duality on the electrical side of "current" (very strongly related to torque in the motor) and "voltage (very strongly related to speed). AC motors can be a little more complex than that, but DC motors are on a first order, exactly like that. Speed correlates to motor voltage, torque relates to motor current.
When a Peltier device is generating electrical power, it is creating an output voltage. And if there is a load, some output current. The thermal part of a Peltier device will have temperature drop across it (think of thermal resistance) as the heat flux goes through it (like electrical current). So heat flux (Q) on the thermal side is related to voltage on the electrical side, and temperature drop on the thermal side (delta T) is related to current on the electrical side.
On the aerospace projects that I've used Peltier devices on, we even used this to modulate the heat flow. Short the electrical side together (no voltage drop) will give you the lowest temperature drop on the thermal side. Opening the electric circuit side gives you the largest temperature drop on the thermal interface part.
But note: present day Peltier junctions are really pretty bad, and are not efficient. For instance, they would not terrible to cool your house, as they exist right now.
But there have been CPU coolers made that you put electrical energy into a Peltier junction to cool the CPU more. Just like a heat pump on your house. -
AgentBirdnest So, it's made by JayzTwoCents?Reply
What an odd photo choice for this article. A billion photos of CPUs, and y'all went with a JTC screengrab. : P -
The Historical Fidelity
I think it represents the technology quite well. By having the thermal transistor in its off state it shows the properties of the physical material used to create it, and the 13x increase in conductivity when it is on proves that the performance is more than the sum of its materials. However, I agree it does nothing to show its effects on a chip’s temperature.TJ Hooker said:The 13x value is the difference in thermal conductance between the thermal transistors being in an 'on' vs 'off' state. Not the difference compared to conventional cooling methods, as incorrectly stated in this article.
I'm not sure why that's an important metric though. At least in the context of cooling computer chips, simply having high thermal conductivity seems like key property, rather than the ability to change conductivity. The IEEE article also highlights the frequency at which the thermal conductivity can be switched, which I also don't understand the significance of.
It’ll be interesting to see how manufacturers choose to utilize this tech. Perhaps they will test out a chip without thermal transistors and determine areas in need and re-design with these thermal transistors in key areas. I hope the tech will be used to normalize the thermal output across the chip, IE: no more hot spots on the chip. -
bit_user
For a chip with a hotspot, you'd better be able to remove that heat if you had a dedicated channel to draw heat away from it, rather than a shared channel that's pulling heat from everywhere else, also. The reason being that a dedicated channel should have a higher temperature delta, and Newton's Law of Cooling tells us the rate of thermal transfer is proportional to the temperature delta.TJ Hooker said:The 13x value is the difference in thermal conductance between the thermal transistors being in an 'on' vs 'off' state. Not the difference compared to conventional cooling methods, as incorrectly stated in this article.
I'm not sure why that's an important metric though. At least in the context of cooling computer chips, simply having high thermal conductivity seems like key property, rather than the ability to change conductivity.
Because a CPU is a dynamic thing, these hotspots surely move around, depending on the code being executed & the data it's processing. If your hotspots move around, then you need a way to dynamically admit/block heat to that channel. That's where I see value in the thermal transistor idea.
If you're going to dynamically reconfigure your thermal channels, then you need to be able to do so about as fast as the hotspots move around.TJ Hooker said:The IEEE article also highlights the frequency at which the thermal conductivity can be switched, which I also don't understand the significance of. -
t3t4 The design innovation that made this possible was a one-molecule-thin layer of molecules that become thermally conductive when charged with electricity
This quote kinda hurts my brain. A molecule of what exactly? For some reason I'm envisioning 'ion wind' through a 1 molecule air gap.
Otherwise this idea sounds a whole lot like a TEC/Peltier which are anything but power efficient! So it can't be that, can it?