Internet enterprises and chips

Internet enterprises and chips
Pingtouge Semiconductor Co., Ltd 2018 year 10 month 31 Established in Japan , Semiconductor company under Alibaba .
Pingtouge semiconductor will be built for automobiles 、 Home appliance 、 Zhilian chip platform in industry and many other industries . Supported by Ali in the early stage , In the later stage, Flathead will operate independently , self-sustaining .

Links to references
https://mp.weixin.qq.com/s/2u6QeCx1Hf8ct6qJHYRE_Q
https://mp.weixin.qq.com/s/ngi3jVdd2xnMG7pj1ojFIQ
https://baijiahao.baidu.com/s?id=1715464505739502565&wfr=spider&for=pc
https://baike.baidu.com/item/%E5%B9%B3%E5%A4%B4%E5%93%A5%E5%8D%8A%E5%AF%BC%E4%BD%93%E6%9C%89%E9%99%90%E5%85%AC%E5%8F%B8/23197541?fr=aladdin
https://www.t-head.cn/?spm=a2ouz.12986968.0.0.7df12be0V0XGhh
2018 year 9 month 19 Japan , Zhang Jianfeng, chief technology officer of Alibaba, said at the yunqi conference that , Alibaba officially announced ： Merge the team of Zhongtian Weida Bodhidharma Academy , Set up Flathead semiconductor chip company .
2018 year 10 month 31 Japan , Pingtouge Semiconductor Co., Ltd. was incorporated .
2018 year 11 month 7 Japan , One is brother Ping tou （ Shanghai ） Semiconductor Technology Co., Ltd. is officially registered , The registered capital 1000 ten thousand , And the place of registration is Zhejiang . The legal representative is liuxiangwen . Shareholder information display , Its shareholder sponsor is Alibaba Dharma center （ Hangzhou ） Technology Co., Ltd .
2019 year 7 month 25 Japan , Pingtouge, a semiconductor company under Alibaba, released the first achievement after the establishment of the company based on RISC-V The processor of IP Nuclear iron 910, Qixiaoning, vice president of Alibaba group, said , It can be used to design and manufacture high performance on-chip , be applied to 5G、 Artificial intelligence and autonomous driving . Using this processor, the chip performance can be more than doubled , At the same time, the chip cost is reduced by more than half .
2019 year 9 month 25 Japan , Alibaba is in 2019 The cloud habitat conference officially released the light 800AI chip , Hanguang is one of the three ancient magic swords , Contain without revealing , Light without shine .
2021 year 10 month 19 Japan , Alibaba is in 2021 At the cloud habitat conference in, a low-power 、 UHF RFID Tag chip —— Feather array 600.
2021 year 10 month 19 Japan , Alibaba is in 2021 The first product was released at the cloud habitat conference in 128 Core self-developed cloud servers CPU chip —— Handheld device maker e-ten 710.
stay 2021 Tencent digital ecology Conference , Tencent officially disclosed its actions in chip for the first time ： in the light of AI Calculated “ Zixiao ” chip , For video processing “ sea ” Chip and high performance network oriented “ Xuanling ” chip , Have made substantial progress . among ,“ Zixiao ” At present, the film has been successfully streamed and lit up smoothly .
Tencent from QQ scratch , Since its establishment, it has been expanding around the software ecology , Stepping into the chip industry , from “ soft ” To “ hard ”, It can be called an important exploration of business layout .
There are two problems behind Tencent's chip making , One is why you want to do , The other is why we can do .
First of all to see “ What do you want to be ” The problem of .
Vice president of Tencent 、 Cloud and smart industry business group COO Qiuyuepeng, President of Tencent cloud, introduced at the conference that , With the development of cloud computing industry , The trend of software and hardware integration is more and more obvious , Software defines hardware unstoppable . Chip is the core part of hardware . Tencent has been actively exploring in the chip field , And has a long-term investment . and AI Calculation 、 Video processing 、 The three scenarios of high-performance networks have strong demands .
Tang Daosheng also responded to the media group interview . Tencent's investment in this area has demand driven factors , Also because some problems need to be solved , Or the need to reduce costs , Or use Tencent's infrastructure more efficiently .
“ such as AI chip ,AI The scale of the service is very large , How to reduce energy consumption through self-developed chips , How to make the cost better , These are all places to promote continuous investment in this field . Another example , Because there are more and more audio and video usage scenarios , As an underlying technology , As a more natural interaction mode for people to people communication , Applied in many scenarios of Tencent , To meet medical 、 education 、 The demand for live broadcasting , It has also become an important place to consume computing resources . In audio and video coding technology 、 Hardware chips have also made a lot of investment , I hope the software and hardware can be integrated , Support audio and video codec more efficiently . Another example is the Xingxinghai server , It's built for the cloud , Multi core chips , Meet the needs of cloud application scenarios .” Tang Daosheng said .
Sullivan and Li Qing, director of cheetah Research Institute, are accepting 《 The daily economic news 》 The reporter said in an interview , The integration of software and hardware is AI The trend of , The support of the algorithm must depend on the capability of the underlying chip , And the existing ones are not enough to match perfectly AI Application of the algorithm model . It's not just Tencent , Huawei 、 Ali is doing AI chip , This is what we are doing AI Platform enterprises are bound to go in the direction .
Look again. “ Do ” The problem of .
Everybody knows , The threshold of the chip is very high . But compared with general chips , More opportunities for dedicated chips , The entry point chosen by Tencent , Special chip .
Tencent's choice is wise , This part of the chip is not open to all walks of life , It has a special purpose , from AI In terms of algorithm application , It also tends to be professional , Video processing , Audio and video technology is what Tencent is good at , Based on this part of the ability to match the combination AI, It is very consistent with the current development strategy . R & D follows the actual demand , Don't follow the trend blindly , No concept , This is very practical .
Of course , Tencent's chip making is not without challenges . Li Qing said , After all, Tencent doesn't start with hardware , The key is the degree of self-study , The degree of cooperation with the outside world . Establishing ecology has always been the principle that Tencent adheres to , This part will certainly be open to partners in the future .
The core of Tencent

Afterthought , Chip verification engineer Lynda Entering Tencent is a bit “ hasty ” 了 .
“ Simulation tools ？”
“ No, , Still talking .”

“ What does the verification environment say ？”
“ Not yet .”

“ that …… The verification process ？”
“ either .“

That is 2019 year 1 month . Join Tencent ,Lynda Ready to roll up your sleeves , Do a lot of work , But face to face “ One question and three questions ”.
“ Is to make chips from scratch .”
During the interview , Leading chip design Henry Gave her a psychological shot . Lynda Just want to participate in the whole process of chip research and development , I'm not afraid to start from scratch . But she really didn't expect , Even simulation tools 、 Verification environment 、 These must-have items in the validation process can “ 3 without ”. as everyone knows , High threshold for chip development 、 High input 、 High risk , The verification engineer is the goalkeeper . The software development can be continuously improved in the later stage bug, If the design problem of the chip is not found by the preliminary verification , Once streamed , Can only be reduced to one piece “ The brick ”. therefore , Many chip companies have a ratio of design engineers to verification engineers 1：3. And in Penglai Laboratory ,Lynda There is only one validation colleague who works side by side , And a line of code without validation conditions . Now ,Lynda Just began to understand Henry In the mouth “ Starting from scratch ” What does it mean , And what a difficult battle she faces .
1
A team that almost disintegrated
Vice president of Tencent cloud 、 According to xieming, general manager of Cloud Architecture Platform Department , The battle of chips , tencent “ A long time ago 、 Be imperative ”. This team is responsible for QQ And other national level applications , Standing at the front line of Tencent's massive business data scouring , Deeply understand the amplification value of the bottom technology innovation to the upper application .2013 year ,QQ Photo album has developed into the largest storage business of Tencent . Let users access photo albums faster 、 The experience is smoother , It has become a very urgent demand . Into corresponding technical problems , Is whether the picture can be transcoded faster ？ Can I compress without compromising the image quality ？ Can you store at a lower cost ？ Ask repeatedly . With the endless self transcendence of software architecture , The team is acutely aware of , Only in the hardware also make innovation , To achieve a deeper breakthrough .

A team with software background , How to make hardware ？
After a round of research , Decided to take FPGA（ Programmable array logic ） Test the waters .FPGA Compared with the chip, the fault tolerance rate is high , But in terms of throughput 、 Delay 、 Power consumption and flexibility are balanced . Especially when dealing with massive amounts of data ,FPGA comparison GPU Significant advantages of ultra-low delay .
Facts have proved this judgment .2015 year , The team concentrated on the research and development of image coding FPGA, It's better than CPU Coding and software coding have higher compression rate and lower delay , It also helps QQ Photo albums greatly reduce storage costs . See in FPGA Direction exploration and in-depth possibilities . 2016 Around the year , from AlphaGo Detonated AI The upsurge has made FPGA Into the mainstream view . The team passed FPGA Of the deep learning model CNN After the algorithm is accelerated , The processing performance has reached universal CPU Of 4 times , And the unit cost is just 1/3. FPGA Although the effect is good , But the technical threshold is relatively high ,“ If you put FPGA Clouding , Is it a solution path that can expand the application ？” With this expectation ,2017 year 1 month 20 Japan , Tencent cloud launched the first in China FPGA Cloud server , Hope to use cloud computing , take FPGA Ability to extend to more enterprises .
In terms of effect , stay FPGA On the ECS FPGA Hardware programming enterprise , It can really improve the performance to general purpose CPU Server's 30 More than times , And just pay the equivalent of ge CPU about 40% The cost of . Take a famous gene testing company as an example , Traditional use CPU We need to test the gene sequence for a week ,FPGA It can be compressed to several hours . However , The cruel reality poured cold water on the team from head to toe , After clouding FPGA, Failed to sweep the entire industry as quickly as expected .
One side , Many enterprises are still not competent on their own FPGA Development , Higher level services are required ; On the other hand ,GPU The rapid decline in costs , Also give FPGA Caused a huge impact . The commercialization of the cloud has been frustrated , The team's enthusiasm from the peak to the bottom , At the same time, it also throws a problem naked in front of the whole team ：FPGA What is the value to the business ？FPGA Can we continue to do ？ Hit by this , Team in 2018 Years are almost falling apart , People began to leave in a centralized way .
2
One good thing came out of ,“ Penglai ” available
Time goes in 2019 year . That was the first year of large-scale application of artificial intelligence , Both internal and external businesses have put forward suggestions on AI The strong appeal of the chip . stay FPGA After the frustration of ECS , Tencent needs to rethink how to continue on the hardware road .AI chip , Do you want to do ？ When the question was raised , There have been dissenting voices from the management , Worried that the technicians are just hot headed , Just to chase hot spots , Just a blind conceit that seems confident , But there is no explicit prohibition of small team exploration . On a small scale 、 Low cost 、 Test the water first in a specific application scenario , It has become a consensus . The Cloud Architecture Platform Department will finalize the first chip AI Reasoning direction , The name “ Penglai ”, I hope this chip can be like the overseas fairy mountain in the ancient Chinese myth , Standing firmly on the surging waves . This hardware breakout team , Also officially named “ Penglai laboratory ”.

experienced FPGA, Penglai lab is quite proficient in hardware programming language , Also in the standard interface 、 Bus and other areas have accumulated some platform design . However , R & D requirements for both , It cannot be said in the same breath . If you say do FPGA Is to build ready-made building blocks , To make a chip is to make blocks directly from logs .FPGA If something goes wrong, it can be reprogrammed , And the chip has only one chance to stream , Once the error , cast to waste . Besides , FPGA Our resources are fixed , The resources of a chip are defined by itself .
One word , Is to “ To dig ”： Do the biggest thing with the smallest resources . Chip architecture engineer Rick use “ decorate ” Change “ The reconstruction ” To describe the whole Penglai project .
In limine , The team thought they could put the previous FPGA The technology is relatively simple to convert into a chip . Doing and discovering , After all, I just think ——FPGA Not many architectures can be directly reused in a chip , The team can only tear down the original structure , The amount of rewritten code is as high as 85%.
image DDR Memory is a top priority , Chip manufacturers usually have special verification personnel responsible for , But Penglai laboratory, which just started, doesn't have this condition , We can only make up for our lessons by taking time .
2020 year 1 month , Penglai chip streaming is completed , Express delivery to Shenzhen by the partner . COVID-19 has just broken out nationwide , The company has started collective telecommuting .

● The Penglai chip completed by the streaming was sent back to Shenzhen by express Henry Take it to the express with gloves , After careful disinfection with alcohol , To an empty office building , Keep the windows and fans wide open , Start lighting operation with several colleagues .
The so-called lighting , Power on the chip , First, check whether there is short circuit and smoke , The next step is to test some basic functions .
Is it a chip or “ The brick ”, Success or failure depends on it .

● Debug the clock of Penglai chip result , The clock frequency of the chip never comes out . Need to know , The clock frequency is of the chip “ Metronome ”, No clock frequency , Different modules of the chip are not aligned properly , You can't work together .
Is it the problem with this chip ？ The experimenter changed a chip , There is still no signal output . Change another piece , Or not? . The scene was silent . The experimenters have been afraid to start . Some people can't help joking , Is it time to go home and change your resume ？ But people are more confused ： Although there are few people 、 Less resources 、 A self-made man , But Penglai laboratory from designers to verifiers , Have confidence to say , Every step is done . What's wrong with it ？ In a very dignified atmosphere , Place the board 、 Access to electricity 、 Read the signal …… The fourth chip , bright . All the chips left , No problem . The truth is so simple .28 Chips for nanotechnology , The defect rate is only 3%, But the first three pieces of the random test all caught up , Let's “ Give birth to a child ” I have experienced a lot of tension . Experienced the twists and turns worthy of the name , Tencent's first chip , Announced .
3
On one layer ,“ Zixiao ” Lingyun
The actual combat performance of Penglai chip has also lived up to expectations , Help Tencent launch the first intelligent microscope approved for clinical application in hospitals in China , Realize automatic recognition of medical images 、 Count the number of cells and display them directly on the visual field , The performance fully meets the design requirements .

● Penglai chip is debugged on the intelligent microscope Effectively create a face-to-face application 、 High performance chips , This road , Tencent can walk . The advent of terminal chip Penglai , Just finished from 0 To 1 The task of . The team can't wait to start 1 To N, March towards large-scale cloud chips . Director of Penglai Laboratory Alex The application for project approval of large chips is called “A Round of funding ”.
After the first try , The team needs to explain to the company , Why do we need to invest more in large-scale chips ？ Can we keep leading in the short and long term ？ How to combine with internal and external business to create value ？ The decision Tencent faces this time , It's much easier to do . Penglai's success has given us enough confidence and confidence . Penglai laboratory grows up while walking , Completed one transformation after another , Build a complete 、 rigorous 、 Standardized chip R & D system and process .
This is already a “ Regular army ”. what's more , The team proved the advantages and position of Tencent as a chip . Xie Ming explained , From the perspective of industry , In addition to considering technology and process, making chips , The biggest difficulty lies in the chip “ Definition ”. The advantage of traditional chip manufacturers lies in the former , But after the chip is made, it will match the demand , In many scenarios, real performance is lost .Google、 The advantage of technology enterprises like Tencent is that they are the demand side , The most profound understanding and insight into requirements 、 Most thoroughly . There is no problem with the direction , Technology and workmanship are no problem , Senior executive vice president of Tencent 、TEG（ Technical engineering division ） President Lushan gave full support , And won more through the general office headcount And money . With the support of the company's strategy , The team goes to a bigger battlefield with high aspirations . Deputy director of Penglai Laboratory Austin It was decided to divide the army in two ways , stay AI Reasoning and video codec are advanced in parallel .AI The team continued to work in Penglai 2.0 edition “ Zixiao ”. This is a 《 The Canonization of the Gods 》 The name of the palace where Li Hongjun lived . Build a prison on the stable fairy mountain “ Zixiao ”, Represents a new ambition ： This time, , Set the goal as No. 1 in the industry .

All the architectures of Zixiao focus on effective computing power . The team optimized the on-chip cache design , And abandon the common GDDR6 Memory , Use advanced 2.5D Packaging technology , hold HBM2e Memory and AI The chips are sealed together , Thus, the memory bandwidth is increased by nearly 40%. Technology iterations take place with each passing day . After Zixiao established the project , The highest performance in the industry has been refreshed by competitive products . Although the design performance of Zixiao is enough compared with the highest performance “ Security ”, But the team is going to continue to overweight . After study , Computer vision is added inside the chip CV Accelerator and video codec accelerator , Innovatively significantly reduce AI Chips and x86 CPU Interaction and waiting between . Even if two complex self-study modules are added , The team is still planning 6 Completed the whole process from architecture determination to verification and streaming within months .2021 year 9 month 10 Japan , Zixiao lights up smoothly .

In picture and video processing 、 natural language processing 、 Search recommendation and other application scenarios , This chip breaks the bottleneck that restricts the exertion of computing power , Finally, in the actual business scenario, the performance has reached the... Of the industry standard 2 times .
4
Independent research ,“ sea ” A smile
2022 year 3 month 5 Japan ,Derick And led the video codec team to receive the streamer back to the chip “ sea ”, At the same time, Shenzhen is fully telecommuting due to the epidemic . Apply for special permission to enter the empty office building . This is the scene , How similar to when Penglai was lit two years ago .
Never thought of , The twists and turns when lighting Penglai , Also reproduce . It overcomes some unexpected problems in debugging , In the cheers , Tencent's third chip 、 At the same time, it is also the first chip independently developed, which has been successfully lit up . From Penglai to Zixiao to Canghai , from 28 Nano technology to 12 Nanotechnology , from 8 Personal development to 100 Many people , From nothing in the simulation tool to “ Tianjian verification platform ” Officially completed , From trying to keep up with the pace of partners to doing it independently SOC.
Penglai laboratory , Completed a scene “ core ” Path Evolution . Besides , It is different from Penglai and Zixiao in application direction AI, Canghai is a video transcoding chip . if QQ The transcoding of photo album pictures is the earliest opportunity for Penglai laboratory to make hardware , The sea came out , It is the completion of a response to the original heart . Multimedia services have evolved from the era of pictures to the era of audio and video live broadcasting ,4K/8K Ultra high definition digital content continues to impact the cloud computing infrastructure . Each additional bit of data , Will bring the corresponding transcoding computing power and CDN Bandwidth cost . This is an intuitive and severe mathematical problem . The goal of the Canghai team is very clear , That is to make one of the strongest video transcoding chips in the industry , Bring the compression ratio to the extreme . Fortunately , Tencent's rich multimedia application scenarios , And many live interactive head customers covered by Tencent cloud , It provides unique analysis and verification conditions for Canghai's research and development .

The team first launched the core self-study module of Canghai —— Hardware video encoder “ The Jade Pool ”, And decided to give yaochi a big test before Canghai finished its research and development . The big test is 2020 Year of MSU World codec competition , By Moscow State University （MSU） The host , It has been one of the most influential top events in the field of video compression in the world for more than ten years , Attracted, including Intel 、 Ying Wei Da 、 Google 、 Huawei 、 Well known technology enterprises at home and abroad, including Alibaba and Tencent, participated in . The result is , Yaochi realizes [email protected] Real time video coding , I won by crushing all the heroes SSIM ( Structural similarity )、PSNR( Peak signal to noise ratio ) and VMAF( Video multi method evaluation fusion ) And other objective indicators , And the human eye's subjective evaluation of the first good results , It is one position ahead of the second place . After this tough battle , Canghai has been fully reviewed technically . Turn the sea into a drop . Canghai will eventually achieve a smaller amount of data 、 Smaller bandwidth provides the same quality of video , The compression ratio is higher than the best performance in the industry 30% above .
5
“100G” Time , The two trees are towering
In Multimedia 、AI While dealing with positive change , The underlying ECS also faces similar problems ： When the performance improvement brought by software optimization can not make the product have obvious competitiveness different from competitive products , How to break through the existing ceiling ？ Another department of Tencent , We have focused on hardware / software collaboration and hardware acceleration .2019 year , Tencent has ushered in a milestone in cloud computing business —— The scale of ECS has broken through 100 ten thousand .
Vice president of Tencent cloud 、 Zouxianneng, general manager of Tencent Network Platform Department, keenly observed , As the server access bandwidth continues to increase , The server is used for network processing CPU There are more and more resources . Whether the server network processing can be realized in a lower cost way , It also provides higher network performance ？ In the face of this “ Both 、 And ” The challenge of , Zouxianeng decides to subtract from the server ：“ Reduce the burden of network data processing from CPU Uninstall it .”“ Smart network card ” The idea was born .
● Tencent smart network card “ Silver fir ”
The so-called smart network card , On the one hand, it shoulders the external network access of the server like an ordinary network card , Realize network interconnection between different servers and data centers . On the other hand , Extra with CPU/FPGA/ Memory and other intelligent units , It can share the virtualization computing tasks of some servers , Accelerate the overall network and storage performance of the server . let me put it another way , What the network platform department should do , Is to install a new server in the network card .
In limine , The team hopes to find a ready-made commercial board to reduce the workload . The person in charge of network card hardware Hayden Take the lead in scheme demonstration and research , However, the acceleration engine of commercial chips did not support private agreements, which became the first challenge at that time , And the biggest obstacle . Some famous network card equipment manufacturers shook their heads at Tencent's request ：“ Now the function of the network card is very simple , This requirement is too complicated , Hard to achieve .” There are also some straightforward doubts ：“ There are so many network cards , High reliability requirements , Can you handle it by yourself ？” Is the smart network card project just starting to miscarry ？ Zouxianneng pointed out the direction for the team ：“ Since the smart network card is a key component of the cloud data center in pursuit of extreme performance and cost , If there is no product on the market that meets Tencent's needs , Then make one yourself .” When the direction is clear , The route soon became clear ： Start with... Based FPGA Self developed intelligent network card started , Then carry out the research and development of intelligent network card chip .2020 year 9 month , Tencent's first generation is based on FPGA The self-developed intelligent network card was officially launched , Name it “ metasequoia ”, The team hopes that the product can be as adaptable as this rare tree 、 Rapid growth . During the epidemic, all kinds of sudden demands hit , The newborn Metasequoia has not been challenged .Hayden Recollection , A major customer has adopted UDP Audio and video protocol , In attribute, it is “ unreliable ”、 Allow packet loss , Heavily dependent on network throughput and stability , Requires high concurrency 、 High quality audio and video transmission effect . Metasequoia smart network card meets difficulties , By greatly improving the network performance of the server , Helped the customer complete 24 Hour zero packet loss limit pressure test , Stable online operation , Handed in a beautiful answer . After Metasequoia glyptostroboides is put into use , Second generation intelligent network card “ Silver fir ” Our R & D work has also started in earnest , And in 2021 year 10 Officially launched in . The network ports of this generation of intelligent network cards have doubled , Reached 2*100G. Under the support of another towering tree , Tencent cloud has launched the industry's first self-developed sixth generation 100G Cloud server .
Calculate the maximum performance improvement 220%、 Maximum storage performance improvement 100%. Compared with the previous generation, the bandwidth of the single node access network has been greatly improved 4 times , Delay descent 50%.
● “ Silver fir ” Carried in the self-developed 100G On the ECS
“ Two trees ” Great gains in network hardware offload , Make the team excited . When FPGA The route is approaching the bottleneck of performance and power consumption , The network platform department decided to take the initiative in its own hands again .
6
“ Xuanling ” At first sight , The core matter is not finished
Tencent's fourth chip , It is also the first intelligent network card chip . There is also a “ It's full of Fairy Spirit ” Name ——“ Xuanling ”, According to the plan , This kind of 7 Nano technology chip will be in 2022 The end of the year .Hayden He was appointed to quickly set up a Xuanling chip R & D team , Constantly challenge multiple “mission impossible”. From the performance index point of view , The number of Xuanling support devices will be increased to 10K above , Compared with commercial chips, it has improved 6 times . meanwhile , Performance can also be improved compared with commercial chips 4 times , By running the original on the host CPU Virtualization on 、 The Internet / Storage IO And other functions are unloaded to the chip , The host can be realized CPU Of 0 Occupy . This short and shrewd chip , Fully interprets the ultimate performance for the future “ Mysterious ”, And the flexible and accelerated... Oriented to various business needs “ spirit ”. at present , The Xuanling project is working intensively to verify and test the intelligent network card before streaming , Build Tencent cloud's next generation high-performance network infrastructure ; Penglai laboratory AI Reasoning chip Zixiao and video transcoding chip Canghai will be mass produced , Deeply integrated with Tencent business ; There are also some new chip projects brewing , Continue to explore the technical direction in need , Enrich this book “ And seas ”. New challenges faced by Tencent's massive business , And the inevitable requirement of the rapid development of cloud computing ,“ Backward force ” Tencent has embarked on this road of making cores . These chips based on business requirements , It must go deep into practical application to prove its value .“ Not out of nothing 、 Pat your head to make a chip . From the very beginning , Tencent's demand is big enough , Enough to do it .” Lu Shan said . from 2010 From the year onwards , Tencent began to open its digital technology and connectivity in the form of cloud services , Go to this era of digital transformation and upgrading of the industry . Bow to the game , Tencent sees that the deep digital real integration is leading the technology trend of all real Internet .
The core of Tencent , Finally, it will be echoed in the sea of stars with the fusion of number and reality .
“ bye , The elder brother of the flat head ”,“ bye , Alibaba ”
lately , Peace head brother's friends chat , It is learned that brother Pingtou will probably operate independently . Because there is no official announcement yet , It's possible .
To hear the news , Despite psychological preparation , Still a little surprised .
Brother Pingtou is a typical case of cross-border entry of Internet enterprises into chip enterprises , Started very early .
The Internet is an application layer based on the current communication infrastructure . Originally, it had little to do with the underlying hardware such as chips . Because the whole people are interested in high technology , Emphasis on basic science and technology , Internet companies have also begun to set foot in the chip industry .
Not satisfied with just doing applications , But to do the underlying technology . This pursuit is commendable .
BAT, Whether it's old bat It's new bat, Have been involved in the chip , And Ali is undoubtedly the first to make efforts .
Internet companies make chips , There are advantages . Because you need a lot of servers , Then you can develop server chips for your own use . in addition , It can also be applied to its own algorithms , Make targeted optimization , Make a difference , At the same time, solve the problems at the application level , Many benefits .
There are two sides to everything . If you only optimize for your own application , These chips will also be difficult to sell to third parties , The possibility of selling to dilute costs is greatly reduced .
Chips are high investment , High risk industries . Advanced technology often costs millions or even tens of millions of dollars at a time , And once 5nm project , Even as high as 3 RMB . Even at such a high cost , There is no guarantee of success .
2019 year , Xiaomi is surging S2 Tape-out 6 Secondary failure , It costs tens of millions each time . Reality has directly dispelled Lei Jun's crazy idea of making chips into sand .
In addition to the film costs , And high patent licensing fees ,EDA Tool usage fee .
before , The employment cost accounts for a small proportion of the overall cost of the enterprise . Now? , This is already an average annual salary of hundreds of thousands of individuals , An industry with a very high proportion of millions of annual salary .
Mid last year , Brother Pingtou is still at the top of the competition , The fresh students take the lead in giving 50 Ten thousand annual salary has aroused wide discussion . Such a high employment cost is definitely not negligible even for Alibaba, which is rich and powerful .
Ali's salary structure also includes a large number of shares , Ali's stock has fallen by two-thirds relative to its peak . The stock as salary and the real salary , For enterprises, the cost is different .
If chips want to be profitable , There must be a certain amount , To dilute chip costs . If the quantity is small , High R & D costs , Spreading the cost of streaming to each chip will make the chip too expensive , No market competitiveness .
qualcomm , mediatek , Intel ,AMD, NVIDIA these chip companies , For the global market , Almost all mobile phone companies , Computer companies are all customers . Rely on this amount , Only then can we have enough competitiveness and profits to maintain the high investment in chip research and development .
Apple's cell phone , Computers used to be assembled by buying chips , Now we use our own chips . Because they don't sell to other countries , Limited shipments . therefore , Will find , Apple's cell phone , After the computer adopts its own chip , Instead, the more you buy, the more expensive it becomes .
Apple is a special case , Because the customer is no longer based on cost performance . So some people are willing to pay for Apple's high investment in chips . And companies that can achieve brand strength like apple , There are few in the world .
Domestic Hisilicon was also very good before it was blackmailed by the United States , It relies on Huawei's huge hardware business , To dilute the cost .
Alibaba's main business is e-commerce . Despite the cloud business , It also focuses on hardware , However, compared with Huawei's hardware scale, it is still far away . Ali's own demand is not enough to support the development of Flathead brother .
therefore , For a chip company , Developed a very awesome chip , Be proud of , But the more important question is , Is there any one that can walk in quantity , Profitable chips . For an ambitious enterprise like brother Pingtou , Only oneortwo mobile chips are not enough to support its ambition .
Brother Flathead is going out cpu What about authorization ？ Obviously more unreliable . Well understood. , by ip The best authorized enterprise is arm, however arm When you are famous , The scale is not big . To achieve arm The influence of that year , Flathead brother is just starting .
So far, , Flathead hasn't developed strong enough competitiveness . Once the company operates independently , It means that Ali no longer makes money for brother Ping . Can Flathead find a growth point quickly ？ This is a problem .
Flathead has expanded rapidly in the past two years . But know , It is difficult for a rapidly expanding company to be free from factional strife .
A large number of people , Everyone has a different background . People with similar backgrounds tend to form a small group . If not handled well , Many companies will die in factional strife .
The original Zhongtian micro team in Pingtou , And later GPU Of A The team . It is said that the two are independent of each other , There is competition in internal resources , I don't know how it has been running in recently .
Internal discord is common in rapidly expanding companies .OPPO Invest tens of billions in mobile phone chips . A large number of internal are from Qualcomm ,MTK, Zhan Rui's faction . Many people leave because of the failure in factional struggle .
Different companies have different cultures , They don't agree with each other , They all think they are reasonable , In the end, we can only know through competition .
scholars scorn each other , The engineer is a little better , But the top is not necessarily . If you feel that there is no office politics , It is not enough to explain the hierarchy .
By the way , If you change jobs and occupy a high position , If you feel right ceo Or the chairman is not confident , It's better to make a good evaluation , Can I sit in this seat . Don't be attracted by the superficial scenery .
People at a certain level , Work for a new company , The biggest concern should not be salary , But confidence and trust in the highest level of the company . Caichongxin's salary reduction following Ma Yun is a typical example , May feel visionary , In fact, people at this level , I don't think much of salary . therefore , It is not caichongxin who has foresight , It was Ma Yun who gave this opportunity . Give a high salary , A high position , If the boss can't hire enough people , You will find that you have actually gone into a big hole .
There is a company in Shenzhen , High salary recruits people to lead the team to make chips , Find friends through headhunters , Say annual salary 200 ten thousand , And stocks , As a result, my friend refused directly . He said that he was not familiar with the senior management of the company , awe .
Even if and ceo It's a friend. , But also consider , Can I bear this responsibility after I go , Are you important enough . If the responsibility is optional , Strong substitutability , Don't expect to get high equity and salary .
It's a bit far away from here .
Go back to Flathead . It is said that many people have left now . Lack of confidence after independence . Originally, we were looking at Alibaba's fame . Despite some recent declines , But a starved camel is bigger than a horse .
But once brother Flathead becomes independent , That would be called Zhongtian micro ？ How to polish the team competitiveness as soon as possible , How to build competitive and sustainable products , It is the labor pains that brother Pingtou will face after weaning .
For Alibaba , No more blood transfusion for brother Flathead , Financial pressure will drop . Chips, no matter how fashionable , After all, for some companies , It's just icing on the cake , What's available or not . When faced with difficulties , Stop the loss in time although it hurts , But it is wise .
The inflection point of chip supply and demand will come soon , Is to continue to burn money to develop chips , I will still shop around , Purchase more competitive products in the market , Ali seems to have made a choice .
Brother Flathead left Ali , If weaning is successful , It may also open a new Xintiandi , go further .
Similarly, there is the Hongmeng system , Only after leaving Huawei's open source Hongmeng , Only in this way can the manufacturers feel free to use . Can gradually form their own Ecology .
Last , It is hoped that the domestic chips, systems and ecology can develop and grow as soon as possible .

Links to references
https://mp.weixin.qq.com/s/2u6QeCx1Hf8ct6qJHYRE_Q
https://mp.weixin.qq.com/s/ngi3jVdd2xnMG7pj1ojFIQ
https://baijiahao.baidu.com/s?id=1715464505739502565&wfr=spider&for=pc
https://baike.baidu.com/item/%E5%B9%B3%E5%A4%B4%E5%93%A5%E5%8D%8A%E5%AF%BC%E4%BD%93%E6%9C%89%E9%99%90%E5%85%AC%E5%8F%B8/23197541?fr=aladdin
https://www.t-head.cn/?spm=a2ouz.12986968.0.0.7df12be0V0XGhh