Qualcomm has been scolded for two years, whose fault is the fire dragon? -Qualcomm, Chips, Fever–Fast Technology (Media of Drive Home)–Technology changes the future

In the past two years, friends who want to change to Android must have heard such a rant:

“888, 8Gen1, X do not buy!”

Although this is a joke, it can be seen how “cool” Android’s flagship products have been tossing everyone in the past two years.

The “strongest performance” at the conference can be heard every year, but the power consumption can’t stop.

Let’s move the time back three years.

At that time, no one would brag about how many good cooling black technologies they used. .

But there is no way, everyone is scolded, and if you want to change to an Android phone, there is no other choice.

However, I still have to ask a question.

Why is Qualcomm so hot in the past two years?

Fever problem is inevitable

We know that as long as work is done, it will definitely bring fever.

For the chip, this heat is mainly composed of dynamic power consumption and static power consumption.

Dynamic power consumption(Dynamic Power) mainly refers to the heat generated by the chip during operation, including the charging and discharging of the circuit, and the jump of the working state of the transistor.

Static power(Static Power) mainly refers to various types of leakage currents and competing currents in the chip.

Taking a switch as an example, dynamic power consumption is like the power consumption generated when we switch the switch repeatedly.

The static power consumption is like the power consumption of the switch in order to maintain its current state (whether it is on or off).

What about the chips we make? . .

Naturally, it is hoped that the density of transistors will become higher and higher, the performance of the corresponding chips will become stronger and stronger, and the power consumption will become lower and lower.

However, the chip doesn’t tell you what “ideal” is.

In recent years, we have made the “switch” smaller and smaller, and the dynamic power consumption has indeed been reduced a lot.

However, as the chip design entered the nano field, the leakage problem of static power consumption began to roll over and became more and more serious.

Attributable to the reason, it can be understood that the “switch” is too thin to stop the electronic “smuggling” on both sides. . .

So if you want to reduce this leakage, you need a whole new structure and new materials.

Rebuild this thin, yet “stowaway” switch.

When the chip reaches 28nm, this technology is FinFET (fin field effect transistor).

Now that the process is slowly reaching 5nm, FinFET is not very useful.

The leakage level is like a circle, returning to the predicament of the 28nm process.

At this time, the big guys came up with a new technology in the future to solve the problem, called GAA (Gate All Around Transistor).

The difference between these two processes and the previous ones is simply to expand from two-dimensional to three-dimensional.

The contact range is increased, and the leakage current can be better controlled.

However, probably for conservative and process validation reasons, no one has used GAA yet.

So there is no choice at present, everyone’s 5nm can only continue to use the old process FinFET.

Repair the seam to hard against this leakage problem.

Isn’t it the craftsmanship?

Then since everyone is hard against this leakage, why can the Apple next door be able to control the energy consumption so well?

This has to promote the well-known first pot man:

Unlike Apple, Qualcomm’s two generations of SoCs are produced by Samsung.

Don’t look at Samsung, TSMC calls their technology 5nm.

But in today’s FinFET era, the name of the process does not correspond directly to the physical parameters of the chip, it is more like the name of a technology node.

The transistor density gap made by the two 5nm processes can even be plugged into an Intel 14nm. In terms of transistor density alone, TSMC’s 5nm process can place 17.13 million transistors per cubic millimeter, or 171.3 MTR.

However, Samsung’s 5nm can only achieve the level of 126.5MTR.

Here I mention a little extra knowledge. The reason for these differences in transistor density is related to the technical iteration routes of the two manufacturers.

Let’s make a long story short, for TSMC, we can understand that TSMC’s 7nm -> 5nm is a complete technology iteration route.

Samsung on the other side is a lot more radical. In their plan, maybe 7nm -> 3nm is a complete technical iteration. They plan to launch a new GAA directly on the 3nm process.

So for Samsung, its 5nm process is slightly conservative compared to its own 7nm progress.

But to be honest, from the perspective of a single transistor, this gap is actually not bad.

Others think that the yield rate of Samsung’s process is also worth talking about.

Two months ago, a news made Tony laugh. . .

No, why is your yield rate so low, you don’t even know it yourself? ? ?

According to @wccftech reports, the yield rate of Samsung’s foundry department on the 4nm process is only 35%.

The yield rate of Samsung’s 3nm GAA technology, which is still under development, is only 10% to 20%.

According to DigiTimes reports, Samsung has falsely reported the yield rate on 5nm, 4nm, and 3nm processes.

We don’t even know whether the current yield rate is the figure before the lie or after the lie.

I don’t know if this is the reason, but Qualcomm decided to hand over the 4nm 8 Gen1 plus to TSMC for foundry.

Oh, by the way, what is the yield rate of TSMC’s 4nm process next door?

70%. . Of course, we are analyzing the process and density gap here, and we can only say that it is a side-by-side reasoning about the gap between Samsung and TSMC’s process.

But in Tony’s view, the “black pot” of this heat cannot be thrown to Samsung’s foundry.

Or the structure can’t stand it?

Why does Tony think it’s wrong to blame Samsung for everything?

Because Dimensity 9000 is coming.

The Dimensity 9000, which we once had high hopes for and was handed over to TSMC, is not very ideal in power consumption.

Data source, Geek Bay▼

In some test scenarios, it can’t even beat his little brother Dimensity 8100.

so. . . Could the problem be with the ARM public version architecture they use?

Tony will analyze it for you. The architectures of the Snapdragon 8Gen1 and Dimensity 9000 are very consistent, and the 1 + 3 + 4 structure of the ARM public version is used.

That is, 1 X2 large core + 3 A710 large cores + 4 A510 small cores.

Just look at the top performance mentioned at the press conference, but what level are these cores heating up?

It can only be said that it was appalling.

In the Geekbench 5 test, the power consumption of the large-core A710 can run to 2.1w, while the single-core power consumption of the ultra-large-core X2 can even exceed 4w.

A single core!

You know, the generation of God U of that year, the whole CPU of Snapdragon 865 can only run to 6.7w under the test. . .

Although it is said that the performance of 8Gen1 and Dimensity 9000 has improved, but this nearly doubled heat is not something that ordinary mobile phones can handle.

What Tony has known in the past two years, the mobile phone that can suppress this heat dissipation looks like this:

Yes, you have to plug the fan inside.

According to the tests of other media, when 8Gen1 is fully exerted, the peak power consumption of CPU and GPU can both exceed 10W!

Apple next door also has a chip that can break through 10w in the past two years. You might as well guess what it is?

It is the M1 on the mobile side.

The same 10w power consumption, what can you do with M1. . . What can this 8Gen1 do. . .

“God is making 8Gen1″▼

And there is another situation. Now ARM is optimized for 64-bit applications. In the case of high load, the super-large core X2 can be used, and the small-core A510 is prepared in the case of low load.

Therefore, when ARM runs 64-bit applications, it needs performance and performance, and power consumption and power consumption.

But because the Android camp has not eradicated 32-bit applications, ARM only retains the large-core A710 to run 32-bit programs.

Therefore, once an Android phone runs a 32-bit application, no matter the size of the application load, it must be thrown on the large-core A710 to run. Even if the app is a notepad, the A710 has to run.

This result. . . No wonder it’s not hot.

Apple is different. It has been researching its own architecture since A7, and it has been updated to A15 this year.

It is completely different from the public version of ARM.

Moreover, Apple is also a ruthless character, throwing away the burden of 32-bit applications in 2017.

So it seems that Apple’s overall power consumption control is better.

It’s hot, now what?

After taking this inventory, we can understand why the high-end chips on the Android side are getting so hot.

Process technology with more and more serious leakage + Samsung’s 5nm “water injection” + ARM public version architecture design is too aggressive + ARM has sacrificed for compatibility with 32-bit applications.

These sets of lying dragon and phoenix chicks just got together.

The final result is that consumers are dumb to eat Coptis chinensis. The phone is really hot. . .

Fortunately, the manufacturers are not the ones who don’t know how to change.

After all, on the way to explore new craftsmanship, who can say that he will never roll over?

TSMC was jokingly called “Taiwan leakage” because of process problems when it was 28nm.

And what Samsung is betting on is naturally the planned 3nm process GAA. According to industry sources, the GAA process can even pull the process to the equivalent of 1nm.

For mobile phone manufacturers, the “rejection” of 32-bit applications is becoming more and more obvious. The action is also getting faster.

This year, when many mobile phone built-in app stores put on the shelves, they also began to force developers to upload 32-bit and 64-bit applications at the same time. Everyone uses 64-bit applications, and the phenomenon of big core waste is gone.

Now, Qualcomm has acquired Nuvia, a startup founded by former Apple chief architect Gerard Williams, for $1.4 billion.

Although it is still some distance away from rebuilding its own architecture self-research team.

But I also hope to use their team’s R&D experience to reduce their dependence on the ARM architecture. It is not impossible for Qualcomm’s own chips to catch up with Apple one day.

By the way, in the next few days, it will be Qualcomm’s new product launch.

The products brought by that time, even if it is not the 8 Gen1 Plus that everyone is thinking about, are mostly new chips manufactured by TSMC.

Although everyone is generally not optimistic about ARM’s architecture this time.

But it is uncertain, this time TSMC has performed exceptionally well, can it bring us surprises?

Pictures, sources:

Some pictures are from the Internet

Part of the data comes from Geek Bay

Why is Samsung’s 4nm inferior to TSMC’s 4nm?

The Cultivation Road of Digital IC Back-end Design Engineer |

https://www.qualcomm.com/news/releases/2021/01/13/qualcomm-acquire-nuvia

https://en.wikipedia.org/en-mo/FinFET

https://m.eprice.com.tw/mobile/talk/102/5717005/1/

In-depth analysis | Why did 5nm chips collectively roll over? The problem that plagued TSMC and Samsung 10 years ago is back – xinhehui

Hashtag: Qualcomm chip fever