To revive Moore's Law, Intel decided to connect chips with glass

In the past two years, the theory that Moore's Law is dead has been widely spread.

But don't worry about nailing the coffin board, because according to the recent news, it may be alive again...

What resurrects Moore's Law is not a rare thing, but the glass that we can use every day.

Recently, Intel released news on its official website that the next generation of advanced packaging substrates, they intend to replace organic materials with glass.

The reason is not that glass is cheaper or better-looking, but that they found that chips that use glass as a substrate perform much better than organic materials.

More intuitively, using glass as a chip substrate has two benefits:

One is to improve the efficiency of signal transmission in the chip, and the other is to significantly increase the density of the chip, which in turn leads to better performance.

This is good news in the wild growth of large models and the shortage of computing power.

Intel officials also released a bold statement, saying that before 2030, they can expand to 1 trillion transistors on a package.

Shichao turned over the graph of Moore's Law, the current limit of a package of transistors is 134 billion, from Apple's M2 Ultra chip, 1 trillion data is directly nearly 10 times compared with it.

And then on the graph, it's quite in line with Moore's Law...

Seeing this, I guess you may make such a muttering in your heart, glass is not a rare material, it really has such great ability?

Before answering this question, we need to understand the basics of chip substrates.

The chip substrate is the protagonist of the last step of packaging, which is used to fix the wafer (die) cut from the wafer in the previous step, and the more wafers are fixed on the substrate, the more transistors the entire chip will naturally have.

For example, the entire packaged chip is equivalent to a city, if the substrate chip is a skyscraper, then the substrate is equivalent to the public transportation that connects these buildings in series, and the transistors are the people who live in the building.

There are only two ways to make transistors, that is, the entire city, more people:

One is to do urban planning under the existing public transportation resources, corresponding to the chip package is to improve the process.

The other is to build more and taller buildings, provided that the city's public transportation system must be fully upgraded, and the corresponding material of the substrate is changed.

Of course, in the process of chip packaging development, these two methods alternate.

From the beginning of the 70s to the present, the chip substrate material has gone through two iterations, the first chip substrate relies on lead frames to hold the wafer.

Intel 4004 chip

Intel 4004 chip baseboard

In the 90s of the twentieth century, ceramic substrates gradually replaced the previous metal lead frames because of better sealing and good thermal conductivity, and then in the 00s, the most common organic material substrates we now have appeared.

Compared with ceramic substrates, organic substrates do not need to be sintered, which is less difficult to process and is also conducive to high-speed signal transmission.

So far, organic material substrates have been regarded as the vanguard in the field of chips.

But organic materials also have disadvantages, that is, the thermal expansion coefficient between it and the two materials of the wafer is too different.

It's okay if the temperature is low, but as long as the temperature is slightly too high, one has a large degree of deformation, and the other is small, the connection between the wafer and the substrate will be broken.

Isn't the chip burned out...

Therefore, in order to avoid this situation, the size of the organic substrate is generally not too large.

The size is small, but if you want to have more transistors on it, you have to work the process, and for this reason, manufacturers in the industry have also used eighteen martial arts.

From focusing on planar packaging to later began to engage in stacking music, that is, stacked packaging.

In the field of stacked packaging, it is now also rolled out of the sky, after many iterations, has come to the most advanced through-silicon via technology (TSV), that is, let silicon chips stack, and then vias to connect.

But now, no matter how advanced packaging technology is, they have begun to stretch thin in the face of the development trend of Moore's Law.

Take TSV technology, for example, although it can multiply the number of transistors to a certain extent, but at the same time its technical requirements are higher, not to mention the cost.

Furthermore, the requirements for next-generation packaging technology are package sizes exceeding 120 mm* 120 mm.

As mentioned above, since the organic substrate is composed of a material similar to synthetic resin, it is easy to bend when heated.

Now the packaging design of the chip requires the chips to be put together one by one, heating is certainly unavoidable, and organic materials for larger package sizes are definitely no fun.

This knife has already been put on the head of the organic substrate, anyway, this life will be changed sooner or later.

How to reform, by whom?

We already gave the answer at the beginning - glass. The glass here does not mean that pure glass is used as a substrate, but the composite resin-like material in the previous substrate is replaced with glass, and the metal edge is still there, similar to the figure below.

Of course, glass is not the kind of glass we use every day, but will be adjusted to create a glass with properties close to silicon.

Compared with the previous organic materials, the glass replaced this time mainly focuses on its three properties: mechanical properties, thermal stability and electrical properties.

The first is the mechanical properties, the glass substrate in the mechanical strength of this piece is the lifting organic substrate.

When the glass is used as a substrate material, holes are made on it to ensure the transmission of signals.

Because glass is super flat and easier to lithography or encapsulate, the number of holes in it is much greater than on organic materials in the same area.

It is equivalent to building more dense public transportation on glass than on organic materials, and there will be more lines.

According to Intel, the spacing between vias at the glass core can be less than 100 microns, which directly increases the interconnection density between wafers by a factor of 10.

The interconnect density increases, and the number of transistors that can be accommodated in the same area is also higher.

Then there is thermal stability, and the glass substrate is not easy to warp or deform due to high temperature.

In case there is a special case, the glass also contains silica, and the properties of silicon are close, and their thermal expansion coefficient is similar, even if the temperature is too high, the chip on the substrate and the substrate are deformed together at the same expansion rate.

Finally, the unique electrical properties of the glass core, more precisely, is actually the electrical performance of the glass after the opening, its dielectric loss will be lower, allowing clearer signal and power transmission.

In this way, the power loss in the signal transmission process will be reduced, and the overall efficiency of the chip will naturally be raised.

The combination of these performances is reflected in the final chip that if it is packaged with a glass core substrate, the number of chips that can be placed is 50% more than other chips.

However, there is another problem, since the performance of glass substrates is so good compared to organic substrates, why not use glass substrates earlier?

In fact, it's not that you don't want to use it, but you want to replace a material, it's not such a simple thing, early exploration, mid-term research and development, and late landing, which is all about throwing money and time.

Take Intel, for example, which began to develop glass core substrates ten years ago, and the funds thrown into it are less than a billion dollars.

The result now is that a set of test tools has been assembled, and the actual mass production of glass core substrates will have to wait until 2026 or later.

Of course, not only Intel, but also many companies in the entire industry are engaged in the research and development of glass substrates, after all, the replacement of organic materials by glass is also a consensus in the industry.

For example, more than half a year ago, Japan's DNP also revealed that it was developing glass substrates to replace traditional resin substrates, and they also set a small goal: to win 5 billion yen in sales by 2027.

To say that the earliest to enter the glass substrate, it has to be SKC subsidiary Absolics, even last year, it has invested 600 million US dollars, planning to build a factory in Covington, Georgia.

According to their plan, it is no surprise that by the end of this year, a small batch of glass substrates began to be produced.

Of course, in a short period of time, the mainstream of the chip substrate market will still be organic materials, after all, the technical iteration to complete the commercialization transformation also needs a transition period, technical cost, yield, etc. are all problems that manufacturers need to solve.

However, it is certain that the importance of organic materials on the stage of chip substrates will gradually be replaced by glass.