A Thermoelectric Cat Warmer
from Microprocessor Waste Heat
Simha Sethumadhavan
Doug Burger
Department of Computer Sciences
The University of Texas at Austin
Motivation
• Hot laptops
• Cold cats
– Frozen whiskers
– Reduced pest control
Solution
Purr
Heat
This talk
On chip
Thermoelectric
Generator
Current
Thermoelectricity
• Thermoelectricity: Electricity produced from heat
• First observed by Seebeck in 1822
Hot End
V = S.T
TH
Wire
i
Thomas
Seebeck
Cold End
Tc
Replica of
the apparatus
Traditional Uses
Seiko “Thermic” watches
5°C body heat, 60W
Doped Poly Si, .3% efficiency
Cassini space probe
32.8Kg radioactive plutonium
fuel, InGaAs thermocouple,
628 Watts, 3-4% efficiency
Cat Mutator
Docile
Cat
Radioactive
Plutonium Pellet
The Physics
e Electrons: current flow
p Phonons: heat flow
p
e
e
p
e
p
e
e
e
p
e
p
e
e
p
Hot end
Cold end
When a wire is heated electrons and phonons diffuse
• Electrons
– Higher electron diffusion  more current (good)
• Phonons
– Collide with other phonons and increase heat flow (bad) or
– Either transfer their momentum to electrons (good) or
– Lose their momentum due to boundary collisions (good)
Traditional Materials
Ideally for large thermoelectric current
• Low phonon flow
– Const temperature difference  Low thermal conductivity
• Many high energy electrons
– Small resistance  High electrical conductivity
• Many phonon electron collisions
– Large voltage per unit temperature difference  High
Seebeck constant
Constant
Metals
Insulators
Semiconductors
Seebeck
Small
High
Acceptable
Electrical
High
Very Low
Variable
Thermal
High
X
MediumHigh
Nanotech allows constants be controlled independently & precisely
New Thin-film Wires
p
e
e
p
e
p
e
e
e
p
e
p
e
e
p
Cold end
Hot end
Thin film (few nanometers)
• Thin film and metal boundary do not align
– More phonon boundary collisions
– More electron phonon collisions
• Figure of Merit (M =
seebeck2. elec/therm)
– Traditional Poly Si is 0.4
– Thin-film Bismuth Telluride is 2.38
–
[Venkatasubramanium et al. Nature 2001]
Generator Efficiency
Maximum theoretical
efficiency of any generator
Efficiency


Th - Tc  1 M 1 
=
 

T
c
Th
 1 M  

Th 
Chip temperatures

• Cold end (Tc)
Temperature
Difference
• Hot end (TH)
Max. efficiency of a
Bismuth Telluride
Generator
50
7.1%
25
3.7%
– 27°C
– 77° C, 52 ° C
• M for Bismuth Telluride
– 6x better
Horizontal Generator
Hot end
Horizontal Generator
(nanowire bundles)
Cold end
Wiring
Layers
Die
• Run a bundle of Bismuth Telluride nanowires
from processor hot spot to cold spot
• Temperature difference: ~50 degrees
Vertical Generator
Wiring
Layers
Hot surface
Cold surface
Die
Vertical
Generator
• Run a bundle of Bismuth Telluride nanowires
from logic level to the heat spreader
• Temperature difference: ~20 degrees
Multiple Generators
Vertical
Generator
Cold surface
Hot surface
Die
Purr
Rough Estimates
Parameters
Horizontal
Vertical
Length
1mm
.25mm
Area
300nm x 300nm
1cm x 1cm
Resistance
13M
.3 
Temp Diff
50
25 (50)
Real Power
.13W
.15W (.6W)
Theoretical
7.1W
3.7W
For Bismuth Telluride:
• Seebeck coefficienct 243V/K
• Resistivity: 1.2 x 10-5 ohm/meter
Conclusions
• Limitations
– Manufacturing
– Engineering: Hinders cooling, peripheral circuitry overheads
– Only cats are supported
• Final thoughts
– Thermoelectric heat extraction looks interesting
– Newer materials can improve power output further
– How far can this be pushed?
– When does this become interesting to architects?
Thank You!