这是今年9月发布在网站http://credibletarget.net/上的一篇文章，由化名“优述”、常住北京的老外看中国集成电路的崛起。基于Tech Insights拆解的最新5G智能手机Mate 50 Pro 5G并分析了基站的半导体组成成分。
在手机端，除个别模拟器件（华为的MIPI和Cirrus的放大器）/射频前端器件（高通）之外，基本上没有其它美国货了。一些严重依赖美国半导体企业的器件，也都转向了日本/韩国的供应商。
更多的信息可以阅读英文原文：
What doesn’t kill you, makes you stronger. Mild strokes aside, this is often true. Certainly, Huawei seems to be doing more than OK. And more broadly, looking across China’s semi sector, things are booming. That’s what the Q3 results suggest at least. Far from slowing Huawei down or making Beijing reconsider its “Manufacturing 2025” ambitions, the results reek of US firms being designed out of the supply chain, an accelerated roll-out 5G plan and domestic tech advancement.
Some will no doubt say that “Yeah, we knew China wanted to develop its own semi industry, so what’s the rumpus?” This observation misses the mark. Before private firms were happy with the Americans, and state firms would just tell their bosses they were no good alternatives. But now orders are being pushed towards domestic rivals, even where they are not very good, providing them with revenues today, and confidence about future revenues, with which to fund R&D.
Let’s start with Huawei and their smartphone business.
Its new 5G phone, the Mate 30 Pro 5G, started shipping this month. Thanks to a teardown by Tech Insights here, we can look inside! There’s not much American left.
There’s a couple of MIPI switches from Texas Instruments (TI) and a noise amplifier from Cirrus, both analog components (analog silicon is really hard, harder than digital). There’s also a Front-end module from Qualcomm.
But most of what you see inside is non-US:
Murata, a Japanese firm, seems to be Huawei’s chosen partner for filters and the modules you put them in, pushing out Skyworks and Quorvo.
The DRAM memory inside the Kirin 990 SoC module is courtesy of Korea’s SK Hynix (not Micron).
And look at all the HiSilicon inside: power management and power amplifiers (MB/HB is hard!), RF transceivers, and LNA/RF switches.
And the substitution success is even clearer when you look at Huawei’s other recent, 4G models. It even seems to be figured out how to do its own front-end modules with Murata, which is really not easy.
Meanwhile, Huawei’s domestic 5G infrastructure build-out seems to be going OK too.
For the base stations which supply the radio access network (RAN) in each location, you need a baseband unit (BBU) and 3+ active antenna units (AAU).
The baseband unit A BBU has multiple chips in it: an FPGA, which is the brain (and unlike a CPU is configurable, so you can adjust it as standards change), a DSP (digital signal processor), a modem, switches etc. Huawei was using Xilinx FPGAs, but now it has its own ASIC. This may be a little tricky as the 5G standards are not yet tied down but given Huawei is playing a large role in setting those standards, perhaps their ASIC is designed with knowledge of what’s likely coming.
Huawei has its own 5G transceiver chips, taped out at TSMC on 65nm, but they are low density only (8T8R), which means they can’t handle fast data. They’ve no doubt been building up inventory of the higher density chips too (from, for example, Lumentum). As such, like the ASIC, we’re probably looking at ‘good enough’ tech, rather than the best. If that is indeed the case, and putting aside security concerns, then Huawei’s build-out internationally may be handicapped as other telco operators, particularly in OECD markets, will likely want the best kit.
The active antenna units The AAU comprises the antennae and remote radio unit (RRU). The RRU is for frequency modulation. It needs RF PA, filters and converters. Again, the phones Huawei is now producing suggest it has developed its own in-house analogy chip capability much faster than most people thought (analog chips are much harder to build than digital ones). So maybe it is now able to build its own AAU too.
It is now clear that Huawei and their seniors made a critical strategic decision when the Entity Listing happened: They doubled down on 5G - and told the supply chain to get ready. And it looks so far that the bet is paying off. TSMC, the Taiwan foundry, which manufactures HiSilicon’s chips, announced a 40% increase in capex plans for 2020-21. Put that down to China Inc. telling them what they wanted to buy over the next two year. (And given Beijing’s worries over Washington asking TSMC to not work for Huawei, making sure their financial future was in Huawei’s hands is clever.)
Across the street, analysts have been raising their 5G smartphone sales estimates for 2020-21 by great leaps. One boutique research firm has Huawei at 75mn units in 2020, 290mn in total. This is big. Apple is only releasing its 5G smartphone in H2. Huawei will only gain more market share from them in China in 2020.
Now, it just comes down to the Chinese consumer to do his/her bit and buy 5G phones. So far, Beijing has refused to say whether it will offer any subsidies for smartphone buyers; maybe they’re waiting to see how sales go.
And in other news… It is not just the smartphone space which has gone full-blown crazy for independence. You can see the same across the semi space.
SMIC, China’s leading foundry, seems to be doing fine: it has announced it is starting scale production at 14nm using its FinFET technology, a first for China. Here’s a good overview, in Chinese. SMIC has always been the also-ran, falling further and further behind TSMC over the years, especially at 28nm. But there does seem to be a greater confidence there these days. They’re still no threat to TSMC and probably won’t make much on 14nm, but that’s not the point. Even our old friends Huahong (my piece here) is making progress, saying it’s aiming for 14nm FinFET production in 2020.
NAURA, China’s only viable semi-conductor equipment company, lodged a nice 20 orders for equipment in Q4, some RMB 500mn worth. Among the clients were YMTC out in Wuhan, which is now building out for mass production of its 64-layer 256Gb NAND chip. YMTC says its ready to produce NAND at scale next year too. It’ll be behind Micron, SK Hynix and Samsung in technology (they’ll be at 128-layer), of course, but they’l no doubt find buyers at the low-end (things like firms who put USB stick together). Not much money to be made, but, hey, that’s not the point in the short-term, as I explained in this piece on China’s disruptive industrial policy here).
