[alert type=”warning” icon-size=”normal”]All of the images below are by Huawei/HiSilicon unless stated otherwise.[/alert]
Let’s start off with an intro here. The name that we’re all familiar with is HiSilicon Kirin when it comes to Huawei devices. Huawei owns HiSilicon, and they design the architecture of Kirin chipsets. HiSilicon is established since 1991, and it’s the 6th fabless IC company in the world. By fabless we mean they’re like Qualcomm – they only design the layout of how the circuit is to be laid out and printed on the chips, but sends the design elsewhere to have it fabricated.
HiSilicon doesn’t only do system-on-a-chip (SoC) designs, but more or less every other electronic device used by Huawei, such as Turing processor, wireless terminal and network, optoelectronic modules, networking devices, and digital media.
During Huawei’s very own Kirin 95X Series Chipset Workshop last week, they proudly presented that the Kirin 95X series has 3 main improvements – higher performance, longer battery life, and improved image processing. We here at Nasi Lemak Tech knows that by the subject of semiconductors and how it affects you and me, can be very confusing and overwhelming. So, we’re here to make it as simple as possible, and explain it to you.
We’ll be breaking down the subject into a few parts, with each part relating to how it contributes to the 3 main claims that Huawei boasted. The parts consist of what HiSilicon does for their Kirin chipset, and how it affects users like you and me.
[nextpage title=”Transistor design”]Fundamentally, all of our devices work because of transistors – an electronic switch that doesn’t have any physical moving parts – which is a type of solid-state electronic. Here, we’ll be specifically talking about the n-channel field effect transistor, or FET for short, for the sake of simplicity. Ever since transistors were fabricated into chips, it looks pretty much like this:
Everyone used this design for make microchips and everything tiny. The main point here though is the part labelled as “gate”. To simplify things, the gate has to be powered to let electricity flow from the source to drain terminal, turning the device on. However tri-gates are mostly used on processors for now.
When transistors become smaller and smaller, the gate length becomes smaller, and the reliability of these transistors decreases. Simply speaking, its reliability is directly affected by the surface area of the gate and the source-drain channel. This is due to the limitation of what is now dubbed as 2D transistors.
A few years back, Intel touted that they will be using 3D transistors, or tri-gate transistor, for their Ivy Bridge-based processors. It was huge. News were flying out everywhere, and Moore’s law somehow managed to (once again) survive a little longer. To be honest, I was really stoked about the technology too.
3D transistors had some radical changes, including the usage of z-axis, or depth, to fabricate transistors. The name tri-gate is a direct reference into the 3D transistor’s gate covering 3 different surfaces of the source-drain channel. This of course greatly improved the contact surface area between source-drain channel and the gate, which drastically increased reliability too. That said, the fabrication process can now be further improved to tiny 16nm FinFET processes.
Intel’s however was for laptops and desktops at that time, but never at the scale of handheld mobile devices. When HiSilicon says that Kirin 950 can have 40% increase in performance and 60% decrease in power consumption, it is all because of FinFET fabrication technique. From the illustration shown above, tri-gate transistors have a desirable side effect where more transistors can be packed closer together to get a high performance increase, which in a way prolongs the life of Moore’s law. As mentioned before, lower power consumption is due to much lesser leakage current between the source-drain channel, thanks to the triple-contact surface gate.
What’s even better is that HiSilicon Kirin 950 uses ARM Mali T880 GPU, which doubles the performance and GFLOPS each compared to Kirin 930.
On a positive side note, tri-gate transistor means improved battery life, lower TDP (thermal dissipation power), and higher efficiency too, because the gate doesn’t require as much power to turn the transistor on and off.
Now there are many other 3D transistor designs which are currently in the works as of now. They’re generally referred to as “multigate transistors”. You can find out a lot more about these gate designs here. Who knows, maybe tomorrow we’ll have a porous source-drain channel diffused in the gate!
[nextpage title=”Photography”]I personally think that this is an interesting point that HiSilicon highlighted. When it comes to photography, everyone either thinks of its sensor, lens, and the image format it supports. While that is not wrong, I just think that there is way much more beyond these 3 things.
Firstly, HiSilicon includes their own PrimISP in the Kirin 95X series chipsets. It is a dual 14-bit ISP, or image signal processor, that works at 940MP/s. An image signal processor of this performance translates to amazingly launching and super-quick focusing, together with ultra-high detail picture quality at a simple click of a button, and much more lag-free, responsive filtering effects to try on.
HiSilicon showed us the performance test of their new camera design compared to the iPhone 6S, and dang I’m impressed. HiSilicon managed to include an amazing intelligent colour management to control the exposure, and the bright parts do not overshadow the less-exposed parts in a picture.
All these can of course be processed via software, but whenever these features are hardwired into the chip itself, its processing is much faster. Also, the processing becomes more passive – even saves battery and faster processing too. This of course, leads us to the next section where HiSilicon shines.
[nextpage title=”Architectural design”]The architectural design of a chipset – in its simplest way, is like this:
Above is a really simple layout of the entire HiSilicon Kirin 950 chipset. As you can see, there are certain components that many might not think it’s meant to be on the architecture at all. And you’re most probably right.
For example – if there’s a block labelled as “4K video” on the architecture, then why aren’t there a “2K video” block? From my standpoint, there’s one main reason which I described earlier. Remember that these processing can be hardwired into the hardware itself, or emulated via software instead. This is exceptionally true for media coder-decoders, or codecs for short. Codecs are absolutely crucial when it comes to saving any form of media into a standardized file format, like JPG, MP3, or MP4. These file formats also dictate the file size to a certain degree too, with their underlying compression algorithm.
Video codecs is the perfect example when it comes to a great chipset architectural design. It can be directly hardwired into the chipset itself, or simulated via software. In this case, the HiSilicon Kirin 950 has H.265, the successor of H.264 codec, directly on the chipset itself. A quick explanation on what H.265 codec is – it is also known as High Efficiency Video Coding, or HEVC for short. It is the answer for 4K video production in terms of file size and video quality. With that said, a dedicated “block” is needed to deal with 4K videos with speed. See this page to learn more about codecs.
In another addition to the story, architecture optimization is pretty important to conserve energy usage too. Everything will have to work properly between one another, and the constant battery for “low power consumption but has high performance” is the single toughest goal to achieve. Kirin pretty much nailed it for the Kirin 950 though.
[nextpage title=”Wrap up”]Being a fabless IC company like HiSilicon is not simple. These companies deal with the challenges on how to design an efficient chipset with the best features, while maintaining reliability and efficiency throughout the entire system. Qualcomm did great for the Snapdragon 801 chipset, but somehow the design went wrong for Snapdragon 810 chipsets, causing a big fiasco due to its temperature issues – even after its 3rd revision.
As you probably have realized, these chipsets are multiple individual modules and components coming together and communicating in a way that is efficient in terms of performance and energy, bringing in other features along the way. That is the reason why both chipsets and SoC, or system-on-a-chip, are really important in mobile devices.