Latest / Mechanical Engineering Made Simple / (#27) Electronic Meltdown Avoidance: Mastering Thermal Resistance.

Transcript

(#27) Electronic Meltdown Avoidance: Mastering Thermal Resistance. · Mechanical Engineering Made Simple

• 0:00
  Welcome back, design junkies, to mechanical engineering made
• 0:02
  simple. Great to be back today.
• 0:04
  We're really diving deep into something absolutely critical
• 0:07
  for, well, almost everything electronic these days.
• 0:10
  You mean keeping things cool, Thermal management and the fluid
• 0:14
  mechanics behind it? Exactly.
• 0:15
  It's the dual challenge, right? How electronics generate heat
• 0:19
  and how we as engineers have to design systems, sophisticated
• 0:22
  systems sometimes, to stop them from melting down.
• 0:25
  It's the unsung hero, isn't it? Get it wrong and performance
• 0:28
  tanks are worse. Total failure.
• 0:31
  So our mission today is to unpack this.
• 0:33
  We're looking at the physics, the key calculations.
• 0:36
  Tracking that heat from the tiny silicon chip.
• 0:38
  Yeah, understanding all the barriers it faces, this idea of
• 0:42
  a thermal resistance network. And then figuring out the
• 0:45
  hardware, the pumps, the fans needed to actually move that
• 0:48
  heat away using, it's quite a journey for that energy.
• 0:51
  It really is from micro scale physics to macro scale
• 0:54
  machinery. OK, so let's start the source.
• 0:56
  The heat is generated in the chip.
• 0:58
  How does it even begin to get out?
• 1:00
  What are the basic ways heat moves in these systems?
• 1:04
  Fundamentally you've got 4 modes, operating, conduction,
• 1:07
  convection, phase change, and radiation, often all at once in
• 1:12
  series and parallel. But right at the start inside
• 1:14
  the chip package through the solid materials.
• 1:17
  There, it's almost entirely conduction heat moving through
• 1:20
  solids. OK.
• 1:20
  And engineers like analogies. Is there a good one here?
• 1:23
  Oh definitely. The electrical analogy is
• 1:25
  perfect. Think Ohm's law, temperature
• 1:27
  differences like voltage. Heat flow is like current.
• 1:30
  Exactly, and that lets us define thermal resistance, your Ede
• 1:34
  which behaves just like electrical resistance.
• 1:36
  It tells you how much temperature push you need for a
• 1:39
  certain amount of heat flow. So for a simple block of
• 1:42
  material, there's an equation for this Ed.
• 1:44
  Yep, a straightforward one. For one-dimensional flow, R sub
• 1:47
  CD equals L divided by the product of K and.
• 1:50
  A OK, let's break that down. L is the length the heat
• 1:53
  travels. Right.
• 1:53
  K is the materials thermal conductivity, how well it
• 1:56
  conducts heat. Higher is better.
• 1:58
  And a is the cross-sectional area the heat flows through.
• 2:01
  Bigger area, easier flow. Precisely so.
• 2:04
  Shorter path, better material, bigger area all reduce that
• 2:09
  basic conduction resistance. But like you said, it's rarely
• 2:12
  that simple. In a real device.
• 2:14
  The heat doesn't just flow in a straight line.
• 2:16
  What happens when, say, heat from a tiny chip needs to get
• 2:20
  into a much larger heat spreader?
• 2:23
  Now you're hitting the complex and often dominant resistances
• 2:26
  first is spreading resistance. Spreading spreading so the heat
• 2:30
  spreads out. Shouldn't a bigger area reduce
• 2:32
  resistance? You'd think so because A
• 2:35
  increases, but forcing the heat to flow laterally sideways
• 2:39
  through the material to reach that larger area, that creates
• 2:42
  its own resistance, an additional penalty.
• 2:44
  OK. So it's the cost of changing
• 2:46
  direction essentially? Kind of, yeah.
• 2:49
  It's resistance due to the heat flux lines having to bend and
• 2:51
  spread out. It could be surprisingly
• 2:53
  significant. And what about when two
• 2:55
  different solid parts touch, like the chip touching the heat
• 2:58
  spreader or the spreader touching the heat sink base?
• 3:00
  That's the other big one, Interface resistance or contact
• 3:03
  resistance, wrote this one. This one could be a real killer.
• 3:06
  Why if they're clamped together tightly?
• 3:08
  Because even surface that look perfectly smooth are
• 3:12
  microscopically rough. Think tiny peaks and valleys,
• 3:15
  asperities. Right.
• 3:16
  So they only actually touch at the tips of those peaks.
• 3:18
  Exactly. The real solid to solid contact
• 3:21
  area is incredibly small. Most of the interface is
• 3:24
  actually a tiny gap, usually filled with air.
• 3:26
  And air is a terrible conductor heat.
• 3:28
  Awful. So the heat transfer across that
• 3:30
  interface, the total contact conductance H sub Co has two
• 3:34
  parts, a tiny bit through the solid contact points H sub C and
• 3:38
  the rest try to jump the gap H sub G.
• 3:40
  And that gap part is the problem.
• 3:42
  Often the biggest problem in the whole stack.
• 3:44
  You can have centimeters of copper with less resistance than
• 3:47
  one poorly managed microscopic interface gap.
• 3:50
  It highlights why things like thermal paste or interface pads
• 3:53
  are so critical. Wow.
• 3:55
  OK, so you have the basic conduction resistance, the
• 3:58
  spreading resistance, the contact resistance.
• 4:00
  How do we sum this all up for, say, a whole chip package?
• 4:04
  We use an overall figure of merit, the total thermal
• 4:06
  resistance R sub T. It's usually defined
• 4:09
  empirically. Empirically meaning based on
• 4:11
  measurement. Yeah, R sub T = ( t sub J minus
• 4:16
  T sub FL close parenthesis divided by Q sub C.
• 4:20
  So the junction temperature T sub J minus the coolant fluid
• 4:23
  temperature T sub FL all divided by the chips heat output Q sub C
• 4:26
  basically degrees C rise per Watt.
• 4:29
  Exactly, it tells you the total temperature penalty for getting
• 4:31
  heat from the junction out to the coolant, and it's made-up of
• 4:34
  internal conductive resistances like we discussed.
• 4:37
  Often lumped as R sub JC. Plus the external resistance
• 4:41
  getting heat off the surface. R sub.
• 4:42
  EX. And sometimes the term for the
• 4:44
  fluid heating up as it flows R sub FL.
• 4:47
  Managing all three is key. You can't just focus on one.
• 4:50
  All right, let's move to the external part 4 extensors.
• 4:52
  We've conducted the heat to the outside surface, maybe a heat
• 4:55
  sink. Now we need to get it into the
• 4:57
  fluid, air or liquid. That's convection, right?
• 5:00
  Convection is heat transfer from a surface to a moving fluid.
• 5:03
  The basic relationship is Q = H * S * a difference T sub S -, T
• 5:08
  sub FL. So heat flow Q equals that's the
• 5:11
  heat transfer coefficient. Big One engineers Chase Yeah, S
• 5:14
  is the surface area in contact with the fluid?
• 5:16
  And T sub S -, T sub FL is the temperature difference driving
• 5:20
  the heat transfer. Right.
• 5:21
  So if we define convective resistance or CVD using our
• 5:25
  electrical analogy again. It would be the temperature
• 5:27
  difference divided by the heat flow, so 1 / H * s.
• 5:32
  Precisely or CV equals 1 divided by the product HS.
• 5:36
  Simple concept, but maximizing H&S is where the engineering
• 5:39
  comes in. OK.
• 5:40
  Let's take S 1st, increasing surface area.
• 5:43
  That's why we have heat sinks with all those fins, right?
• 5:45
  Absolutely. Fins or extended surfaces are
• 5:48
  all about dramatically increasing S the wetted area to
• 5:52
  lower George Eva. But is it quite that simple?
• 5:55
  Does a fin work perfectly along its whole length?
• 5:58
  No, good point. The fin itself has internal
• 6:01
  conduction resistance, heat flows from the base out towards
• 6:04
  the tip, and the thin material isn't perfectly conductive.
• 6:07
  So the tip of the fin is cooler than the base.
• 6:09
  Yes, it means the transfers less heat than if it were at the base
• 6:12
  temperature. We quantify this using fin
• 6:15
  efficiency ETA sub F. ETA sub F like a percentage.
• 6:18
  Yeah, it's the actual heat the fin dissipates divided by the
• 6:22
  ideal heat it would dissipate if the whole thin surface was at
• 6:24
  the base temperature. Long thin fins might have quite
• 6:28
  low efficiency. There's a diminishing return.
• 6:30
  Makes sense. OK, so increasing S with fins
• 6:33
  helps, but with limits? What about the other lever
• 6:37
  increasing H, the heat transfer coefficient?
• 6:40
  This feels more complex fluid dynamics.
• 6:42
  It is complex. Fluid flow can be chaotic,
• 6:44
  depends on geometry, fluid properties, velocity.
• 6:47
  It's messy. So how do engineers get a
• 6:49
  predictive handle on H without testing every single
• 6:52
  possibility? We use dimensionless parameters.
• 6:55
  These are clever ratios that group variables together and
• 6:58
  make experimental results more universal.
• 7:00
  They capture the essence of the physics.
• 7:01
  OK, like what? What's the key one for H?
• 7:03
  The most important one for quantifying convective heat
• 7:06
  transfer is the Nusselt number. Nusselt number no town.
• 7:09
  What's it telling us physically? Middle down is essentially the
• 7:12
  ratio of heat transfer by convection compared to what
• 7:15
  you'd get by just pure conduction through the fluid
• 7:17
  layer right next to the surface. So high NERDA beans conduction
• 7:21
  is really working well boosting heat transfer beyond simple
• 7:23
  conduction. Exactly the formula is nu equals
• 7:26
  H * L / K sub FL where L is a characteristic length and K sub
• 7:31
  FL is the fluids thermal conductivity.
• 7:34
  We often use correlations to find neutral based on how the
• 7:37
  fluid is flowing. Which brings us to how the fluid
• 7:39
  is flowing. That's the Reynolds number,
• 7:41
  isn't it? See Absolute classic Yes there's
• 7:43
  one tells you about the flow regime.
• 7:45
  It compares the inertial forces, the tendency of the fluid to
• 7:48
  keep moving, to the viscous forces, the internal friction
• 7:51
  slowing it down. And the formula.
• 7:52
  RE equals rho times V * L divided by MU.
• 7:57
  Rho is density, V is velocity, L is that characteristic length
• 8:01
  again, and MU is the dynamic viscosity.
• 8:04
  And the outcome? What does Striella predict?
• 8:06
  It predicts whether the flow is smooth and orderly, laminar, or
• 8:09
  chaotic and mixed turbulent. Low $2.00 is laminar, High $2.00
• 8:13
  is turbulent. And for cooling, we generally
• 8:16
  want. Often we want turbulent flow
• 8:17
  because all that mixing brings cooler fluid constantly to the
• 8:20
  surface, which dramatically increases H and therefore no
• 8:23
  dull. But turbulence sounds like it
• 8:25
  would need more energy to create more pressure drop.
• 8:28
  Absolutely, that's the crucial trade off.
• 8:30
  Turbulent flow gives you much better heat transfer, but you
• 8:33
  pay for it with significantly higher friction losses.
• 8:36
  This means you need a more powerful fan or pump, using more
• 8:39
  energy, making more noise. So engineers aren't just blindly
• 8:43
  chasing the highest possible turbulence, no.
• 8:45
  Not at all. Good design finds the the sweet
• 8:48
  spot often. Just getting into the turbulent
• 8:50
  regime gives a big boost, but pushing $2.00 higher and higher
• 8:54
  might give diminishing returns on heat transfer while the
• 8:57
  pumping power cost skyrockets. It's always an optimization
• 9:00
  game. OK, one more dimensionless
• 9:02
  number you mentioned earlier for systems without fans or pumps.
• 9:06
  Ah yes, natural convection, where the fluid moves purely
• 9:10
  because heating makes it less dense and buoyant.
• 9:12
  That's governed by the Rayleigh number one, DAR.
• 9:15
  So new $3.00, so new dollars. These are the tools for
• 9:19
  predicting age. They are essential.
• 9:20
  They let us use established correlations from experiments to
• 9:23
  predict performance for new designs.
• 9:25
  So if natural convection isn't enough, which it often isn't for
• 9:29
  say high power processors, we need forced convection.
• 9:33
  We need fans for air, pumps for liquids.
• 9:37
  How do we choose the right one? OK, now we bridge from thermal
• 9:40
  and fluid physics to system hardware.
• 9:43
  The first step is figuring out what the system demands from the
• 9:46
  fluid mover. Meaning how much fluid needs to
• 9:48
  flow and how hard the machine needs to push it.
• 9:51
  Exactly. We need the required flow rate Q
• 9:54
  and the total pressure increase or developed head HS, sub P that
• 9:58
  the pump or fan needs to provide.
• 10:00
  And how do we calculate that required head HP dough?
• 10:03
  It comes from the energy balance on the fluid circuit.
• 10:06
  You start with something like the Bernoulli equation, but you
• 10:08
  have to add terms for all the energy losses.
• 10:10
  Losses due to friction in pipes, bends going through the heat
• 10:13
  sink itself. Precisely.
• 10:15
  We often use the Darcy Weisbach equation or similar methods to
• 10:19
  calculate those frictional head losses.
• 10:21
  They typically depend on the square of the flow rate.
• 10:23
  So the faster you try to push the fluid, the resistance goes
• 10:26
  up sharply. Yes, and when you plot the total
• 10:30
  head required by the system versus the flow rate Q, you get
• 10:33
  the system head curve. The system head curve.
• 10:35
  What does that usually look like?
• 10:36
  It's generally a curve that starts at some value
• 10:39
  representing any static head difference and rises roughly
• 10:42
  parabolically. So H required equals some
• 10:44
  constant K sub one plus another constant K sub 2 * Q ^2.
• 10:48
  OK. So we know the point on that
• 10:50
  curve where we want to operate our desired Q and corresponding
• 10:53
  H. Now how do we pick a machine
• 10:55
  that can deliver that efficiently?
• 10:57
  The key classifying parameter, the sort of figure of merit for
• 11:00
  rotating fluid machines, is specific speed and sub.
• 11:04
  Specific speed nensuf that sounds imortant.
• 11:07
  What is it? It's a nondimensional number
• 11:09
  that relates the machine's optimal operating point, its
• 11:11
  speed, flow, and head to its impeller geometry.
• 11:15
  It tells you what type of machine is best suited for a
• 11:17
  given job. Than the formula.
• 11:18
  N sub S = N * Q to the power of 0.5 all divided by H to the
• 11:23
  power of 0.75. N is the rotational speed.
• 11:26
  OK. What does a high wonder call a
• 11:28
  value mean versus a low one? Is there a rule of thumb?
• 11:31
  Definitely high Nets machines are good for high flow rates but
• 11:34
  low head requirements. Think axial flow fans like a
• 11:39
  desk fan or a propeller. They move a lot of air but don't
• 11:42
  build up much pressure. And lowness.
• 11:44
  Lowness machines are designed for high head, meaning high
• 11:47
  pressure but lower flow rates. These are typically centrifugal
• 11:50
  pumps or fans with more complex impeller shapes designed to
• 11:53
  really throw the fluid outwards and build pressure.
• 11:56
  So calculating the required Q&H then finding the next out guides
• 12:01
  you to the right type of fan or pump.
• 12:03
  It's an incredibly powerful selection tool, gets you in the
• 12:06
  right ballpark immediately. But for liquid systems,
• 12:09
  especially pumps, there's a critical check we absolutely
• 12:11
  cannot ignore, right? Something about boiling?
• 12:13
  Yes, net positive suction head or NPSH.
• 12:17
  This is non negotiable. It's about preventing
• 12:18
  cavitation. Cavitation.
• 12:20
  That sounds bad. It is if the pressure at the
• 12:23
  pump inlet drops too low below the liquids vapor pressure.
• 12:26
  At that temperature, the liquid will locally boil, forming vapor
• 12:30
  bubbles. OK, bubbles form, then what?
• 12:32
  Then, as those bubbles are swept into the higher pressure regions
• 12:35
  inside the pump impeller, they collapse violently implode.
• 12:40
  And that causes damage. Major damage.
• 12:42
  It sounds like gravels going through the pump.
• 12:44
  It erodes the impeller material and drastically reduces
• 12:47
  performance. It can destroy a pump quickly.
• 12:49
  So how do we prevent it? What's the check?
• 12:52
  We compare 2 values. First, is NPSH available or
• 12:56
  NPSHA one? This is determined by your
• 12:58
  system, the static liquid level, the tank pressure friction
• 13:01
  losses in the suction pipe. It's the absolute pressure
• 13:04
  margin above vapor pressure that your system provides at the pump
• 13:07
  inlet. OK, that's what the system
• 13:08
  gives. You right then you compare it to
• 13:10
  NPS required or NPS 01. This is a property of the pump
• 13:14
  itself, determined by its design and provided by the
• 13:16
  manufacturer. It's the minimum pressure margin
• 13:19
  the pump needs at its inlet to operate without cavitating.
• 13:22
  And the rule is simple. Absolutely simple and absolutely
• 13:25
  critical. NPSHA must be greater than NPSH
• 13:28
  R1, usually with some safety margin.
• 13:30
  If it's not, you need to redesign the system inlet, lower
• 13:33
  the pump, or choose a different pump with a lower NPSHR.
• 13:36
  All right, we've selected our pump or fan.
• 13:38
  It's installed, but systems don't always need to run at full
• 13:42
  blast. How do we control the output
• 13:44
  efficiently? You mentioned those system
• 13:46
  curves. Yes, and this is where
• 13:48
  understanding the pump or fan laws, sometimes called affinity
• 13:52
  laws, becomes incredibly useful, especially for saving energy.
• 13:56
  These laws tell us how performance changes if we just
• 13:59
  change the speed. Exactly, assuming we keep the
• 14:01
  same machine, same impeller diameter and just change the
• 14:04
  rotational speed and the relationships are quite elegant.
• 14:08
  OK, what are they? Flow rate Q is directly
• 14:11
  proportional to speed, and so half the speed half the flow.
• 14:15
  Simple enough? What about the pressure?
• 14:16
  The head. Head H is proportional to n ^2
• 14:20
  half the speed, 1/4 of the. Head n ^2 OK and the power
• 14:23
  needed to run it, that must be the important one for costs.
• 14:25
  That's the kicker. Power P is proportional to n ^3.
• 14:29
  Cubed. Wow.
• 14:31
  So even a small reduction in speed gives a huge power saving.
• 14:34
  Massive. If you reduce speed by just 20%
• 14:37
  to .8 N, the power drops to .8 ^3, which is only about 51% of
• 14:42
  the original power. Yeah, you save almost half the
• 14:43
  energy. That explains the big push
• 14:46
  towards variable speed drives VSD's, doesn't it?
• 14:49
  Instead of the old way. Absolutely.
• 14:50
  The old way was often throttling.
• 14:52
  You'd install a pump size for the maximum possible demand and
• 14:55
  when you needed less flow you just close the valve part way.
• 14:58
  Which just wastes energy, right? Forcing the pump to work hard
• 15:01
  against an artificial restriction.
• 15:03
  Exactly, it's like driving your car with the accelerator push
• 15:05
  down and controlling speed with the brake.
• 15:08
  Hugely inefficient VS DS. Instead, just slow the pump or
• 15:12
  fan down electronically. So the machines performance
• 15:15
  curve shifts down to meet the system curve at the lower
• 15:19
  required flow rate naturally. Precisely.
• 15:22
  And because of that N Cube relationship for power, the
• 15:25
  energy savings are enormous. We're talking millions of
• 15:28
  kilowatt hours per year in large industrial or HVAC systems.
• 15:32
  It's a major factor in energy conservation.
• 15:34
  OK, shifting gears slightly, we focused on traditional
• 15:37
  mechanical cooling. Does the source material mention
• 15:40
  any alternative, perhaps solid-state methods?
• 15:44
  Yes, briefly it brings up thermoelectric coolers or techs.
• 15:48
  Techs, how do they work? No moving parts.
• 15:51
  Correct, they use the Peltier effect.
• 15:53
  It's a phenomenon where passing an Electro current through a
• 15:57
  junction of two different semiconductor materials causes
• 16:00
  one side of the junction to absorb heat get cold, and the
• 16:04
  other side to release heat get hot.
• 16:07
  Like a tiny solid-state heat pump.
• 16:09
  That's a perfect description. They're used for spot cooling,
• 16:12
  niche applications, or where reliability and lack of
• 16:15
  vibration are paramount, even if efficiency isn't always the
• 16:18
  best. And how is their performance
• 16:20
  measured? Is there a figure of merit like
• 16:23
  a dollar dollars for pumps? There is.
• 16:25
  It's called the figure of merit, usually denoted by ZZZZ.
• 16:29
  What defines it? The formula is Z equals alpha
• 16:31
  squared divided by the product of rho and K.
• 16:33
  OK, let's unpack that alpha. Alpha is the C Beck coefficient
• 16:37
  related to the voltage generated by a temperature difference.
• 16:39
  It's linked to the Peltia effect.
• 16:41
  Rho is the electrical resistivity of the materials.
• 16:43
  And K is the overall thermal conductance of the device
• 16:46
  structure. Right, so to get a good tech you
• 16:49
  want materials with a high C back coefficient, low electrical
• 16:52
  resistance so you don't generate too much internal heat from the
• 16:55
  current, and low thermal conductivity so he doesn't leak
• 16:59
  back from the hot side to the cold side.
• 17:01
  It's a material science challenge.
• 17:03
  Wow. OK.
• 17:04
  We've really covered the whole chain then from the heat
• 17:07
  starting in the silicon. Battling through conduction
• 17:10
  resistance, the tricky spreading and interface resistances.
• 17:14
  Needing to overcome the convective resistance to get
• 17:16
  into the fluid using our understanding of new on and
• 17:19
  Newell dollars. Which then dictates the need for
• 17:22
  a specific pump or fan chosen using specific speed and checked
• 17:27
  for safety with NPSA check. And finally operated efficiently
• 17:31
  using VS DS thanks to those n ^3 power laws it all links
• 17:35
  together. It really does.
• 17:37
  The whole process is a network, a series of resistances from the
• 17:40
  junction right out to the ambient air or water.
• 17:43
  So maybe a final thought for our listeners, something to really
• 17:46
  chew on. I think it's this because it is
• 17:48
  a chain of resistances, our dollars being the total.
• 17:51
  Simply improving one part in isolation might not help much if
• 17:56
  it wasn't the bottleneck. You mean like putting a
• 17:58
  massively powerful fan on a heat sink with terrible contact
• 18:02
  resistance at its base? Exactly.
• 18:04
  The fan might lower the convective resistance
• 18:06
  dramatically, but if the contact resistance retinal set to was
• 18:09
  already huge and limiting the heat flow to the fins, the
• 18:13
  overall improvement artiliz might be tiny.
• 18:15
  So the real art is identifying the largest resistance in the
• 18:19
  path, the bottleneck, and focusing your engineering effort
• 18:22
  there. That's where you get the most
• 18:23
  bang for your buck. Balancing all the resistances
• 18:26
  from the microscopic interface right up to the fan selection is
• 18:29
  what leads to truly efficient and effective thermal design.
• 18:32
  It's a system problem. Until the next equation design
• 18:35
  junkies, we are mechanical engineering made simple.