The Truth about CPU Soldering
Skylake delidding seems to be very common by now. Every day I read postings from people complaining about Intel and the thermal paste between IHS and die. Even tho Skylake is performing great, people are not satisfied with the temperatures on load. Compared to older generations there is conventional thermal paste between the IHS and the die while Sandy Bridge and older CPUs were soldered. Why did Intel change the production and is the thermal paste really that bad?
Solder what and why?
Nowadays CPUs are produced in the C4 flip chip package (Figure 1). The silicon-wafer is the base to create the CPU. Due to it’s monocrystalline structure it’s possible to create perfect layers on an atomic level. So after creating the integrated circuit on top of the substrate (Figure 1: bottom of the substrate), metalized pads will be placed on the chip as a connection. Solder balls will eventually connect the chip with the PCB.
The die itself will produce a quite large amount of heat compared to its size – that’s one reason why you need an integrated heatspreader (IHS). Another reason is the LGA socket. The Skylake PCB is about 0.78 mm thick and will bend if you only apply pressure on the die (Haswell PCB is around 1.17 mm thick). So you also need the IHS to keep the PCB straight for a perfect connection to the pins in the LGA socket.
The key is how to connect the IHS with the die for a perfect thermal connection. The semiconductor material is usually silicon and the IHS is made out of copper because of its good thermal conductivity of about 400 W/(m*K) – plus it’s reasonably affordable. There are several theoretical ways to connect both materials but in the real world you are limited by a lot of factors such as the maximum temperature the chip can sustain and the thermal conductivity of the joint-material.
Figure 1 shows how a current Intel CPU looks like (Ivy Bridge, Haswell, Skylake…). The substrate is connected via solder bumps with the PCB which eventually connects the CPU with the LGA socket. The underfill is necessary because PCB, substrate and bumps have different thermal expansion coefficients. Thus the underfill is protecting the CPU from destroying itself due to thermal expansion. The heatspreader will conduct the heat from the substrate to the heatsink which will sit on top of the heatspreader. The adhesive is flexible and will alow thermal expansion without damaging the CPU.
Depending on the type of CPU you can either use a normal thermal compound between substrate and heatspreader or connect both with a solder TIM.
How to solder silicon and copper?
You have to understand that silicon (Figure 2) and copper are completely different materials. Silicon (Si) looks like a metal but behaves and feels more like glas (e. g. SiO2). The thermal conductivity is quite good with about 150 W/(m*K) and the thermal expansion is relatively low 2.6 µm/(m*K).Copper (Cu) on the other hand is a ductile metal with very good electrical and thermal conductivity. However, the thermal expansion with 16.5 µm/(m*K) is more than 6 times higher than silicon. You can solder normal copper wires with several tin (Sn) based materials such as Sn60Pb40 (Figure 3), which is a common tin alloy. However, tin based solder won’t stick to silicon at all. In addition, solidifying tin will shrink by a large factor which results in a big thermal tension inside the material. This tension could already damage the substrate of the CPU.
Indium (Figure 4) is the only known material which can stick to both, copper and silicon. At the same time, solidifying indium doesn’t shrink much which leads to a small factor of thermal tension inside the TIM.The thermal conductivity is not as high as copper but higher than any other thermal interface material 81.8 W/(m*K). Common thermal compounds have a conductivity of about 5-10 W/(m*K).
In addition, indium is very ductile which allows thermal expansion of substrate and heatspreader without damaging any of the components involved. Indium melts at 157 °C.
Before we start with the soldering details there are few things you need to know about indium. Similar to aluminium, indium forms a small oxide layer exposed to ambient air. Indium is a post-transition metal and very rare on earth. Compared to the worldwide gold production of about 3000 tons per year, indium is extremely rare. The world production of indium does not even reach 1000 tons per year. The average price of indium was about 800 USD per kilo in 2014. The price dropped now to about 400 USD, but the material is still pretty expensive. Depending on the size of the CPU, the raw indium will already cost about 2-5 USD.
CPU Soldering – The Real Deal
Most people have a normal soldering iron in their mind whenever they talk about soldering. However, soldering DIE and Heatspreader is a complete different story. Indium will stick to both, silicon and copper but the bond strength has a huge influence on the long duration performance and durability. Thus you need to prepare heatspreader and substrate before soldering.
The HeatspreaderThe heatspreader (Figure 5) is covered in a nickel (Ni) layer. Nickel will act as a diffusion barrier to prevent any atoms to form an alloy with the copper. Indium also sticks to Nickel but not really well. So to improve the adhesion you have to apply another layer on top – preferably using a noble metal because they provide the best wetting conditions. Examples would be gold (Au), silver (Ag) or palladium (Pd). The melting point of silver and gold is quite similar at around 1000°C. The melting point of palladium is at 1555 °C which makes it much harder to apply on the heatspreader. So choosing between gold and silver, gold easily forms an alloy with indium and has great wetting conditions. So before you can think of soldering you have to apply a gold layer on the heatspreader. The gold layer has to be around 1-3 µm thick.
As described above, indium is the only material you can use. Depending on the shape of the indium you have to remove the oxide layer before soldering. This can be done by selective etching using hydrochlorid acid.
The indium layer has to be about 1 mm thick to provide enough material for thermal expansion without cracking after few thermal cycles.
The DIE is made out of silicon (Si) but you can’t solder directly to silicon. Otherwise indium would diffuse into the silicon which would result in a different doping characteristic or damage the chip over time. So you need another diffusion barrier on top of the CPU. The diffusion barrier is formed out of several layers made out of Titanium (Ti), Nickel (Ni) and Vanadium (V).
On top of the diffusion barrier you need another gold layer as wetting layer for the indium connection.
The Soldering Process
Let’s take a look at what we have before soldering heatspreader and DIE.
The nickel plated heatspreader is sitting on top. The gold layer underneath the heatspreader will provide the best wetting conditions for the indium. The Indium solder sheet is sitting between heatspreader and DIE. The DIE is covered with 3 layers made out of titanium, nickel+vanadium and gold.
The soldering temperature has to be around 170 °C. Lower might result in a bad soldering connection and too high temperatures will damage the CPU permanently or result in a bad yield rate.
Some of the materials will form alloys during the soldering process. After soldering the result will look like this:
You can see that gold, indium and nickel formed alloys with different thicknesses. DIE and heatspreader are now joined and ready to go.
So far so good. But there are a lot of negative aspects which we will discuss now.
Keep in mind that you also have to glue the IHS to the PCB during the soldering process. Solidifying indium will shrink during the soldering process. As a result, DIE and heatspreader will be pulled together. The result is an uneven heatspreader which you might have read about before.
Typical failure mechanisms
Intense thermal cycling will damage the solder preform significantly. Tensile stress inside the solder preform will lead to voids (Figure 10).
The voids will decrease the thermal conductivity and increase the thermal resistance (Rjc). Eventually micro cracks will occur which usually originate from the DIE corner (Figure 11).
The micro cracks will also decrease the thermal conductivity but will especially increase the thermal resistance at the corner of the DIE. Without the gold layer between diffusion barrier and solder preform, delamination of the solder preform would occure after few thermal cycles. Micro cracks occur after about 200 to 300 thermal cycles. A thermal cycle is performed by going from -55 °C to 125 °C while each temperature is hold for 15 minutes. The micro cracks will grow over time and can damage the CPU permanently if the thermal resistance increases too much or the solder preform cracks completely.
Void and micro crack occurrence is mainly affected by the solder area – thus the DIE size. Small DIE size (below 130 mm²) e. g. Skylake will facilitate the void occurence significantly. However, CPUs with a medium to large DIE size (above 270 mm²) e. g. Haswell-E show no significant increase of micro cracking during thermal cycling (Figure 12).
This failure mechanism is one reason why small DIE CPUs like Haswell-DT or Skylake are not soldered while the large Haswell-EP CPUs are soldered.
I spent months and some $$$ on figuring out how to solder a recent Intel Skylake CPU. After reading all the above you can imagine that it’s by far not as simple as it sounds like. I’m not going to publish all info on Skylake soldering for now because I’m still waiting for the thermal cycling result. So far I can say that it worked great on air.
During Prime95 AVX Load the CPU core temperature is at about 50 °C. The CPU was cooled by a Prolimatech Megahalems with one fan at an ambient temperature of 24 °C. Compared to stock, the temperature dropped by about 18 °C. The result is pretty similar compared to liquid metal thermal compounds. However, liquid metal thermal compounds are not working subzero that’s why I’m trying to find a way to solder Skylake.
Thermal cycling is a significant problem for overclocking with liquid nitrogen because you go from +30 °C to -190 °C. Normal thermal compounds can’t stick properly to the DIE at e. g. -180 °C so extreme overclockers have a lot of trouble getting the TIM to work. The main reason is the different thermal expansion coefficient. Silicon, copper and compound have different coefficients and a sudden heat load can lead to separation of these components.
I sent a soldered 6700K to my friend Splave who is a famous overclocker in the US. Unfortunately the first LN2 test failed. At the moment I’m trying to figure out where exactly the solder preform failed – will keep you updated!
Whenever I read sentences like “What a ripoff – Intel doesn’t even solder a 300 USD CPU” or “Why does intel save 2 USD on soldering” I’m thinking
Stop hating on Intel. Intel has some of the best engineers in the world when it comes to metallurgy. They know exactly what they are doing and the reason for conventional thermal paste in recent desktop CPUs is not as simple as it seems.
Micro cracks in solder preforms can damage the CPU permanently after a certain amount of thermal cycles and time. Conventional thermal paste doesn’t perform as good as the solder preform but it should have a longer durability – especially for small size DIE CPUs.
Thinking about the ecology it makes sense to use conventional thermal paste. Gold and indium are rare and expensive materials. Mining of these materials is complex and in addition it’s polluting.
After soldering one of my 6700K CPUs I can tell it’s a pretty complex process. I’m still working on it and trying to make it available for extreme overclockers. However, I doubt that Intel will come back with soldered “small DIE CPUs”. Skylake works great even with normal thermal paste so I see no reason why Intel should/would change anything here.