Stand in the time node 2016, aurora borealis of turn one’s head is in Yang Lei of chip domainInvestmentLayout, speak of: ” hold water 11 years, our altogether joined 10 companies, among them 3 appear on the market, two are bought exit, 5 all be in harmony reached the rest B annulus or C annulus. In other words, we did not leave miss one gun, 0 error. “

Yang Lei is this held water 2005InvestmentThe trustee of the orgnaizationGeneral manager, before joining aurora borealis 2010, strategy of the technical development that he has 10 one’s remaining years, company andOperation, and investment experience. Look in him, aurora borealis invested a domain to hand over such report in chip, be not stem from accidental good luck.

” if making general chip, we had done Ying Weida very hard. ” Yang Lei says, ” should with Ying Weida, Intel such company competes, we must want to choose a path, take application very greatly, be aimed at applied form to buildHardware. “

Accordingly, each of aurora borealis enterprise that be cast is the angle from a fractionize undertakes be cuttinged, for example dedicated at IoT wirelessRadio frequencyBlueX, intelligence of god of the Cista of dedicated sensor, radium, the Spreadtrum of dedicated processor, OURS, dedicated DSA (Domain Specific Architecture) science and technology of climb a hill, black sesame seed, 100 million wisdom science and technology, the Crossbar of dedicated memory, and the GMEMS that dedicated frequency front handles.

A at the beginning of May afternoon, in boreal capital trade top layer can look down at the aurora borealis of whole townThe officeIn, the chip of Yang Lei and aurora borealis invests a group to invest the logic of backside of these 10 groups a little with us.

Cast only the industry is veteran

” there is a strange elephant in today’s AI chip industry. ” Yang Lei says: ” begin a few years ago, a few enterprises enter this one field, I think they are sexes of the look up before having very much, but these companies are academic group mostly. “

Yang Lei ever was academic group, when he read a doctor 2000, ceng Jin has goneChip design. ” my doctor graduates at that time, it is OK to need a parts of an apparatus to be able to be used only graduation. But want to do trulyProductIf, Fine rateNeed achieves 95% above, this is very difficult. Because a lot of research and development also had been made before me, have blind faith in not at all to academic group so. ” he says

He thinks, the person that academic group does poineering work can drive minor industry person only, can hold independently personally because of top class industry person dish, a stratagem which ensures success is a lot of greater. ” we see company of chip of a lot of AI had reached 100 people now or the dimensions of above, also opened a news briefing, but the news briefing asks afterwards sold how many, do not have sound. ” Yang Lei says.

” these companies are in very inchoate when we have bring into contact with, but did not cast. ” answer relatively with Yang Lei’s observation, 10 of aurora borealis enterprises that be cast come from an industry entirely, have core IP mostly, group experience rises and fall a few times of semiconductor domain too periodic, a lot of have poineering success experience, belong to 2 times do poineering work.

” chip domain, of time it is very important to accumulate, I special the principle that advocates 10 thousand hours. ” Zhao Gu of manager of aurora borealis investment says, before joining aurora borealis, he ever was controller of business affairs of region of China of Intel artificial intelligence, have the government of chip research and development that evens more 14 years, OperationAndThe marketSaleExperience.

Zhao Gu calculated brushstroke Zhang, one has experience capable chip engineer, the education cycle that needs 5 years at least in the industry, probably the investment of millions of RMB volume level. If an institute sends a group to have 100 people, that means a few tuition of 100 millionCost.

” Cong Mofei’s mensurable angle looks, academic group does poineering work, this tuition must want to pay. ” Zhao Gu says, ” this also is the reason that I approbate aurora borealis to invest a concept, investment becomes organizational system company, in the experience of their associate with, tuition has paid. “

Feng Nuo Yi ManFrameworkGeneral no longer applicable

Heat up greatly in AI chip today, the starting point of aurora borealis is not AI chip itself, calculate howeverFrameworkchange. Once was PC or cloud server no matter, the foundation is the X86 framework of Intel. In AI times, rock-bottom foundation has been changed into GPU from CPU, be about to enter melt into for example the other form of TPU.

” ground floor changed, whole modes of life and relation to their environment can produce tremendous change. ” Yang Lei says, ” resembling is a domino same, one poured a chain of thing to be able to produce change, so I spend a long time to think, these changing can happen in where. “

Above all, yang Lei thinks, although current AI computation basically depends on Ying Weida’s GPU, but prospective 2-3 year, we can see a few other companies running. He sees, although the flexibility of Ying Weida chip is unapproachable, but also exist at the same time the issue with higher power comsumption. Although Gu Ge’s TPU is to be aimed at AI task research and development, but outer bound looks, use too hard, too hard.

” real opportunity is between these two extremes. ” Yang Lei says, ” should find voice of a balance, but this balance condition is in after all which position, very rare group can want to be clear about. “

Look in aurora borealis, this ” balance condition ” not be calculative only ” balance condition ” , also be computation and memory union ” balance condition ” . Since the computer is born industrial group is in the framework of Feng Nuo Yi Man that abide by may be issued in future the computational means place of generation artificial intelligence overturns.

” today’s chip does processing partly, one part does memory, but such not differentiating of the cerebra of our mankind. In biology cerebra, computation and memory are to close 2 for of one. ” Yang Lei says, ” a our enterprise that be cast is in what this respect has had a lot of to think. “

The cloud, brim, end, sensor

” does future have Yun Heduan only? ” Yang Lei is cast gave an issue. ” our viewpoint: Either. “

Aurora borealis thinks, stem from answer time, efficiency andSafeconsideration, future can have a few memory and computation to be put in the brim certainly. An any hospitals are center of a data, an any cars also are center of a data, even any family is center of a data, that is to say, the brim changes those who greet world-shaking.

Additional, aurora borealis thinks, in future, be not all terminal unit to be able to become intelligence, the state that can present polarization however — a kind very simple, and another kind of special intelligence.

Simple chip needs to have join, simple calculation and memory only a few basic functions, aurora borealis is in layout of this one domain a company, roll out ” chip namely modular group ” concept. That is to say, the job that ability of a lot of chip finishs needs on modular group today, in future can finish on onefold chip, and this chip is hadCostPower comsumption is low, low, the characteristic with minor volume.

And intelligent chip covers image and sound two kinds of ability, aurora borealis also has corresponding position in these two domains.

On this foundation, aurora borealis thinks sensor will produce very fundamental value, there is relevant investment in the domain such as image sensor, Lidar.

AI chip, ” chip ” proportion is larger

” when everybody is talking AI chip, what more talking is AI, but I think the proportion of chip is larger. ” Zhao Gu says. Aurora borealis thinks, of chipProductionNeed practice and experience accumulate, put in higher threshold.

” is difficulty in? Chip is one is industrial catenary very long, the rigorous industry with complex technological process, any link appears short board cannot achieve anticipateProductWith commercial end. A few companies are inAlgorithmicAfter bonus disappears gradually, begin to look for new topic again, the way that wants to adopt chip goes now, do not have actually so simple. ” Zhao Gu says.

To showing level a fewAlgorithmicThe company begins to grind oneself the phenomenon of chip, aurora borealis thinks, algorithmic company and chip company cooperation will be more desirable means.