Single core A7 plays with face recognition, and NXP "crossover processor" plays a new trick!

6 month 25 Japan ,“ NXP microcontroller business 2019 Media exchange meeting ” Held in Beijing . NXP microcontroller （MCU） General manager of business department Geoff Lees And his team shared their understanding of the microcontroller 、 Views in the fields of edge computing and machine learning , It also introduces the business development strategy of NXP microcontroller and its layout in the Chinese market , And the latest i.MX Processor products .

▲ From left to right , Jin Yujie, marketing director of the microcontroller business unit of NXP Greater China , Senior vice president of NXP and general manager of microcontroller business department Geoff Lees, Vice president of NXP and LPC And Yu Xiujie, general manager of low-power microcontroller product line , Zeng Jintao, global product director of NXP microcontroller division

“ Crossover processor ” Concept recognized

Microcontrollers have always been / Microprocessor （MCU） They are all semiconductor chips that are widely used , Although its performance is very limited , But the power consumption is very low . Thanks to the rise of the Internet of things market in recent years , about MCU Demand for is also exploding . however , With artificial intelligence （AI）、 The rapid development of edge computing and its integration with the Internet of things , Also for MCU Put forward higher requirements . For example, while ensuring low power consumption , Further improve performance 、 energy efficiency , Support edge computing , At the same time, add the ability of machine learning and so on .

This has long been the case 2017 year , NXP semiconductor announced the launch “ Crossover processor ” The concept of , The so-called crossover processor , It refers to a conventional application processor （AP） And MCU Between processors . Its biggest feature is the adoption of MCU kernel , But based on AP Architecture of , Therefore, it can realize AP High performance and rich functions , At the same time, it also has tradition MCU Ease of use （ Continue to use MCU Tool chain ）、 low power consumption 、 Low cost features , Thus breaking through AP and MCU Boundaries between .

and i.MX RT1050 It was the first cross-border processor product launched by NXP at that time , be based on Arm Cortex-M7 kernel , Main frequency up to 600MHz, Can provide 3015 CoreMark/1284 DMIPS Processing speed of , Performance is higher than any other Cortex-M7 product 50% above , Higher than the general market Cortex-A5 product 100% above , It is the highest performance in the market in the same period Cortex-M7 Solution .

And once this product is launched , It was quickly recognized by the market . According to NXP ,RT1050 stay 2017 year 10 Monthly postpartum , In less than a year, it has owned 2500 Customers , The market opportunities found are as high as 5 Billion dollars .

Thanks to the first cross-border processor product i.MX RT1050 Great success in the market ,2018 In, NXP launched two new cross-border processor products ： More powerful RT1060（ Chip memory increased to 1MB、 Support high speed GPIO、CAN-FD And support synchronous parallel interface NAND/NOR /PSRAM Extended storage controller , Graphics acceleration is also added ） And focus on high cost performance RT1021.

▲ The main control of Dajiang's latest mecha master education robot is NXP i.MX RT1021

Besides , At that time, NXP also announced the launch of a product based on 28nm FD-SOI（ Fully depleted insulating silicon ） Process manufacturing support Linux Microprocessor products of the system ——i.MX 7ULP, Not only the performance is greatly improved , At the same time, the power consumption is close to the lowest power consumption of microprocessors on the market . however , At that time, NXP did not elaborate on i.MX 7ULP To introduce .

Incorporated into the “ Crossover processor ” Of i.MX 7ULP

In this year 6 month 25 Japanese “ NXP microcontroller business 2019 Media exchange meeting ” On , NXP analyzed the media in detail for the first time i.MX 7ULP This product , At the same time, this product is also classified as “ Crossover processor ” In the product line .

Yu Xiujie said , At present, there are more people in the market who are interested in “ Crossover processor ” The concept of . Introduced to you two years ago “ Crossover processor ” Concept is not the end , Introduced today i.MX 7ULP It's a new one “ Crossover processor ” Concept .

According to introducing ,i.MX 7ULP For the first time Arm Cortex-A7 kernel + Cortex-M4 Combination of kernels , These two cores are in two completely independent domains , in the majority of cases ,Cortex-A7 The core is closed , It only starts when there is a need for high performance ,Cortex-M4 Deal with some basic tasks more , Ensure low-power operation of the equipment . It's kind of like Arm In the field of mobile phone processors big-LITTLE Big and small nuclear architecture , Its purpose is also to provide powerful performance at the same time , It can further reduce the power consumption . however , take Arm Of AP It seems to be the first time that small cores and microcontroller cores have a similar combination of large and small cores .

And the reason why I chose Cortex-A7 and Cortex-M4 This combination , Mainly because Cortex-A7 Is currently the Arm Highest energy efficiency ratio 、 One of the most widely used AP kernel , At the same time, the cost is also very low , And you can run Linux operating system . Again ,Cortex-M4 It is also widely used in MCU among , Energy efficiency ratio is also very high , The cost is also relatively low . use Cortex-A7 and Cortex-M4 The combination of , It can not only provide the best performance and power consumption , At the same time, it can also have lower cost and higher ease of use .

Besides ,i.MX 7ULP Inside Cortex-A7 The domain , It also integrates low-cost on-chip DDR as well as Vivante GPU kernel （ Original core technology GPU, Has been acquired by Xinyuan microelectronics ）. It also makes i.MX 7ULP Capable of complex graphics processing acceleration .

NXP LPC And Yu Xiujie, general manager of low-power microcontroller product line ,”i.MX 7ULP A good application scenario of is smart watch , Because the smart watch is dormant most of the time , It may take less than half an hour to launch various applications or answer the phone in a day , therefore i.MX 7ULP Very suitable for smart watches , It can provide high-performance support required by various application functions , It can also ensure its ultra-low power operation , To achieve longer endurance “.

We all know ,MCU And application processor in power consumption is very different , Can be said to be 1-2 An order of magnitude difference . such as , be based on MCU Xiaomi bracelet can be used for several weeks after being charged once , While using application processor Apple Watch It can only be used at most 2 God . although , Conventional MCU Low power consumption , But it can't support Apple Watch Requirements for high performance and complex function realization . therefore ,i.MX 7ULP This product is really suitable for smart watches .

Zeng Jintao, global product director of NXP microcontroller business unit, explained , about Apple Watch For a class of smart watches ,” Sometimes it is necessary to deal with complex graphics i.MX 7ULP In the middle of Cortex-A7 open , Close immediately after running . Built in 2D GPU and 3D GPU It's the same way , You can go through it first 3D GPU Deal with the picture , Then close it , Then passed 2D function . For example, it is applied to smart watches , Can pass 3D GPU Process the dial image , The switching time can be through Cortex-M4 perform “. In this way, we can meet the performance and functional requirements at the same time , Minimize power consumption , Greatly extend the battery life of smart watches .

besides , Thanks to the built-in Cortex-A7 Kernel and GPU,i.MX 7ULP It also supports face recognition and intelligent access control 、 Application of face recognition intelligent door lock .

▲ The on-site display is based on i.MX 7ULP Face recognition intelligent access control , Based on reading technology 2D Face recognition algorithm

▲ Another model based on i.MX 7ULP Face recognition intelligent module , It can support infrared camera and infrared fill light mode , Integrated high-precision 3D Face recognition algorithm ; In addition, it can support USB Of 3D Structured light cameras , Integrate 3D Face recognition algorithm . Can support up to 5000 Zhang's local face image comparison and recognition , It can also support the import of online face image database .

Yu Xiujie said ,“ These customers' needs have a feature , Or it needs good display ability , Or you need strong computing power , Or with the network 、 The cloud keeps communicating , But these applications need low power consumption . We think i.MX 7ULP It is the best product for these applications , It has been adopted by many customers ”.

It was revealed that , at present i.MX 7ULP Production has begun , And in a very short time , Has sold more than 100 Thousands of stars , It has been recognized by many customers .

Yu Xiujie emphasized ：“ new i.MX 7ULP, Follow i.MX RT The products in the series are very different . The most important thing is that it has two cores , One is Cortex-A7, The other is Cortex-M4, Its operating system is also different . If we just do MCU Or just be an application processor , It's hard to make this kind of product that appears in our introduction . NXP is the only one in the world MCU And application processors , Therefore, it is natural for us to do cross-border processors , If you only do MCU Or just an application processor should be a cross-border processor , It's very difficult .”

Chinese design + Cross border processors manufactured ：i.MX RT1010

In particular, it is worth mentioning , This time, NXP also launched a cross-border processor designed and manufactured in China ——i.MX RT1010.

▲ At the media conference that day ,Geoff Lees It's a great pleasure to announce ：“ this （i.MX RT1010） We really designed it in China 、 A chip made in China .”

▲ Made by SMIC i.MX RT1010 The wafer

According to introducing ,i.MX RT1010 This product is completely designed by NXP China R & D team , And delivered to SMIC Beijing Yizhuang factory for OEM production , It's using 40 Nano process . It not only has smaller package size , And the main frequency has also been increased to 500MHz, The overall performance has reached twice that of the competitive products , More importantly , Its price is only the original i.MX RT1050 Half of , as long as 6.88 element , about 1 About $ . so to speak , It's very cost-effective .

“ This is not in the traditional sense that you have seen 1 Below $ MCU.i.MX RT1010 The dominant frequency can reach 500MHz, Its computing power is higher than that of Intel's Pentium processor . The price is 1 Dollar products can achieve such performance , This was not available in the market in the past .” Zeng Jintao emphasized ：“ At present, it has reached 500MHz Of MCU, The price is usually as high as 5-6 dollar 、 even to the extent that 8-9 dollar . and 1 Dollar price MCU, The dominant frequency can only reach 100MHz perhaps 100MHz The following levels . The ability to do this at this price ,i.MX RT1010 It's the first one .”

“eIQ” And “EdgeScale”

As mentioned earlier , With artificial intelligence （AI）、 The rapid development of edge computing and its integration with the Internet of things , Also for MCU Put forward higher requirements . For example, while ensuring low power consumption , Further improve performance 、 energy efficiency , Support edge computing , Simultaneous addition AI And so on .

In order to play its own chip AI Ability , This time, NXP also launched a machine learning development environment developed by NXP itself “eIQ”.

Now there are all kinds of AI Model , According to different models, the hardware is also different , It can be used CPU/MCU function , It can also be used. GPU/DSP function , You can also use a dedicated AI Coprocessor to run .

And NXP launched eIQ platform , The machine learning model can be optimized according to different hardware . According to introducing ,eIQ The environment can be compared with almost all open source training tools on the market 、 modeling 、 Editor compatible . There are almost all open standards or customized standards in the market , Can be developed in this development environment . therefore , be based on eIQ platform , Developers can easily deploy machine learning applications to all product lines of NXP , From low-end microcontrollers to high-end processors .

Geoff Lees Express ：“ Our set eIQ Our platform can help customers optimize their algorithms , Customers can choose a hardware suitable for their applications according to their own needs for power consumption or performance , Based on NXP MCU Or high-end processors develop their applications . Not only high-performance processors are most suitable for machine learning , This development platform can help customers in very low-end MCU Can run up to 10 times 、20 Times or higher performance .”

Besides , NXP also launched a software infrastructure EdgeScale, It aims to achieve ML Application to unify data collection in edge devices 、 Management and handling .EdgeScale It can be compared with cloud based AI （AI）／ML Seamless integration of services , It also supports the deployment of cloud training models and reasoning engines on all NXP hardware platforms .

Jin Yujie, marketing director of the microcontroller business unit of NXP Greater China, said ：“EdgeScale Software architecture is an architecture that supports security device management from cloud to end , It can provide more effective machine learning computing ability . Its passage is in PaaS Build a high-quality development tool on the layer , Provide system level support for a large number of operations . At present, some domestic mobile phone manufacturers 、 factory 、 Network operators have been using this platform to optimize their management environment , And strengthen the management of its safety equipment .”

Besides , NXP can also provide customers with complete AI Solution , namely “ One stop service ”.

According to introducing , NXP and domestic including speech recognition 、 Top algorithm companies in face recognition and other aspects , Have a very wide range of cooperation , The projects involved are also very extensive .

“ We cooperate with face recognition algorithm company , Running face recognition technology on a single core processor （ As mentioned above i.MX 7ULP, In fact, it is through single core Cortex-A7 To run the face recognition algorithm ）, It turns out that this is a very large model in the cloud , Now we have successfully moved it to the edge . Besides , The site also showed the high performance based on enzhipu MCU And iFLYTEK's speech recognition algorithm products . These are NXP and its partners , Optimized based on the complete solution of NXP . In the future, we will continue to provide customers with such a complete solution .” Jin Yujie introduced with examples .

Continue to strengthen investment in the Chinese market

As one of the largest microcontroller consumer markets in the world , NXP also attaches great importance to the Chinese market , And is constantly increasing investment in the Chinese market .

In addition to this key release of design in China 、 Cross border processors made in China i.MX RT1010 outside ,Geoff Lees Pointed out that , NXP Suzhou design center is the first chip design and R & D center in China , The founding celebration was just held at the end of last year 20 Anniversary events . NXP also established a laboratory in cooperation with Suzhou University last year , Provide them with sufficient technical support , Provide better training for students .

meanwhile , NXP also expanded its software department in Beijing last year , Many new software engineers are in Beijing . At present, the Department of NXP in Beijing has exceeded 200 Famous engineer . NXP is based on the capabilities held by the original departments in Beijing , Machine learning is added 、 The ability of artificial intelligence and edge computing . Besides , NXP has also set up a laboratory in Beijing , In order to cooperate directly with customers on specific projects .

At the media conference that day ,Geoff Lees It also announced that NXP will cooperate with Tianjin University to build a joint laboratory of the Internet of things , The establishment of the laboratory was also supported by the Tianjin Binhai New Area Government , And plans to officially open the new laboratory next year . Besides ,Geoff Lees Also revealed , Next, we will work with Shanghai 、 Universities in Guangzhou and Shenzhen cooperate .

6 month 26 Japan , TEDA Management Committee and NXP Semiconductor Co., Ltd 《 Investment cooperation agreement of artificial intelligence application demonstration base project 》 Signing of , NXP AI and IOT demonstration projects have officially settled in TEDA . It is reported that so far , The total investment of NXP semiconductor in TEDA has reached about 8.52 Billion dollars .

Geoff Lees It was emphasized many times at the media meeting that day ：“ The Chinese market has always been the strategic focus of NXP , Define products for the Chinese market 、 Design and manufacturing are the themes that NXP has been emphasizing in the past ”.

edit ： Smart core - Ranger sword