Over the past 50 years,the iteration of chip manufacturing process has been rolling forward along Moore’s Law,and continues to enhance the computing power and performance of the chip.Nowadays,whether integrating more and more functional modules on SoC or further improving chip integration under advanced manufacturing process by using chiplet technology,it fully demonstrates that the improvement of chip performance,power consumption and cost can not only depend on the upgrading of manufacturing process,but also need to expand and innovate from different dimensions to continue the”economic benefits”of Moore’s Law.This has made chip design more and more difficult,and the role of IP has become increasingly prominent.It has gradually become the”secret key”for enterprises to seek design differentiation.
A few days ago,at the 10th EEVIA Annual China Hard Technology Media Forum and Industry Chain Research and Innovation Trend Outlook Seminar held in Shenzhen,Huang Yin,senior manager of Imagination product market,said in a speech that the growth of the semiconductor industry’s demand for high-performance SoC has boosted the IP technology market.As the world’s leading semiconductor IP supplier,Imagination is following the market development trend and continues to carry out technological innovation,The evolution of IP technology in its core field reflects the significant demand changes of some popular applications in the future.
Huang Yin,Senior Manager of Imagination Product Market,introduced at the forum how Imagination IP technology can empower future hard core technology innovation
The”inverted pyramid”allows the value of IP providers to be re-examined
With the vigorous development of the server market,the popularization and implementation of the Internet of Things and AIoT applications,and the 5G+artificial intelligence and other technologies enabling the digital transformation of the economy and society,Huang Yin said that IP is also like a butterfly.The process from egg to larva to butterfly reflects this change in the demand of the application market,and promotes the continuous development of technology.More and more semiconductor manufacturers began to consider using external forces,For example,professional IP companies provide complete technical services in a certain field to reduce costs and save time for products to enter the market.
As a proven reusable chip design module,IP has a global market size of about US$5 billion from the perspective of market value.However,it has levered the continuous evolution of the US$600 billion semiconductor industry and enabled the continuous development of the 1.4 trillion electronic information manufacturing industry.Huang Yin added:”According to lPnest data,in the global chip R&D expenditure of more than 60 billion US dollars per year,IP R&D investment only accounts for 6%of about 3.6 billion,but its value and influence far exceed the share measured by money.As the core of the entire semiconductor upstream industry chain,each chip can support more than 200 yuan of social economy,while each IP chip can support 20000 yuan of social and economic value.”
Billions of dollars of IP industry support hundreds of billions of dollars of semiconductor industry
Why is small IP so important.Chip giants have sufficient technology,talent and capital reserves to face the challenges of the huge market,while small and medium-sized companies and start-ups can only focus on certain market segments under the constraints of resources.However,no matter how large or small the company is,it is facing fierce market competition and hopes to form technical barriers and a strong ecosystem to survive better.In the nearly infinite sea of technology,no company can master everything.IP manufacturers can provide mature products of some core technologies with professional technical capabilities and experience in a certain field,assist chip design companies and system manufacturers,speed up the industrialization cycle of technology,improve flexibility and market reaction speed,reduce R&D risks and operating costs,accelerate product launch and increase profits,and help enterprises focus on the development of their core competitiveness,It is also one of the key competitive factors to realize differentiation.
“Imagination technology innovation always focuses on the richness of functions and the optimization of area,performance and power consumption,and is committed to achieving the best energy efficiency ratio and creating core values that cannot be replicated for customers.”Huang Yin further said,”For IP manufacturers,the three dimensions of PPA are the key standards to be considered,and the IP products of Imagination have been widely used in mobile phones,PCs,tablets,cars,desktop servers,data centers and other fields by adjusting the proportion of the three dimensions to achieve the adaptability to different markets.”
Pursuing the best PPA and enabling unique GPU IP innovation through more than 30 years of technical accumulation
From the perspective of market development trend,with the continuous expansion of the scale of data center servers,flexibility and scalability are driving the technological evolution of IP core suppliers.Power consumption,bandwidth and security are the key issues in chip design.How to innovate at the technical level in response to these needs is the focus of the whole industry.In her speech,Huang Yin confidently said that Imagination,based on more than 30 years of experience and technology accumulation in the GPU IP field,has built an industry-leading value barrier with a number of unique patents and core technologies.
Flexible and scalable GPU hardware architecture technology
Imagination’s GPU hardware architecture has very good flexibility and scalability.According to different market and product requirements,single-core GPU can be configured with different specifications from 1PPC,2PPC to 64PPC.According to different requirements of low,medium and high,multiple different image processing capabilities and floating point computing capabilities can be matched.If a larger configuration is required,a loose-coupled multi-core architecture,or even a joint architecture of several multi-core units,can be further provided to meet the requirements of ultra-high performance and ultra-large computing power PCs,servers and other products.
Imagination’s loosely coupled multi-core architecture technology can flexibly switch modes within a single chip.For example,it can work together as a single GPU multi-core for joint rendering and computing when high performance is required;When multiple different services need to be calculated in parallel,it can work independently as multiple single-core GPUs without interference with each other,which is very practical.This loosely coupled technology can even be extended to multiple chips.A multi-core GPU can be reasonably distributed to different die chips,and then connected again through the chiplet technology.It can achieve the same flexible mode switching as a single chip.It can not only render and calculate multi-core GPUs,but also work independently.In today’s rapidly changing market environment,this technology provides chip design companies with high flexibility and scalability without re-flow,and can flexibly turn around to meet market demand.
Block delay rendering technology(TBDR)
In terms of GPU rendering technology,unlike the immediate rendering mode commonly used by desktop GPUs,Imagination was the first in the industry to propose and adopt the Tile-Based Deferred Rendering(TBDR)technology,adding a delay link after vertex shading and rasterization,making full use of the system memory and greatly improving the overall efficiency by rendering only visible pixel segments and abandoning the covered or occluded pixel segments,Provide higher rendering performance and lower requirements on bandwidth and power consumption.TBDR technology has obvious technical advantages in effective utilization of bandwidth,efficient distribution and processing of workload,which is its uniqueness.
Blocked delayed rendering(TBDR)technology of Imagination
IMGIC compression technology
In order to save bandwidth,most SoC chips in the industry currently use hardware compression technology.According to Huang Yin,IMGIC compression technology can support both completely lossless compression and lossy compression of pixels at the same time,and lossy compression supports four compression levels,including lossy compression with compression rates of 75%(near perfect quality),50%(visual lossless),37.5%(certain loss),and 25%(the most bandwidth saving).After lossless compression,there is no difference with the original image,and the average bandwidth can be saved by about 50%.For lossy compression,the bandwidth and image storage size can be compressed to the specified ratio according to the configured compression level to effectively save.In most cases,it is difficult to detect the difference between compressed images with the naked eye.
GPU hardware virtualization
It is worth mentioning that Imagination is also the first company in the industry to realize GPU hardware virtualization IP authorization,and has been widely used in many models.Huang Yin explained that in automotive applications,using hardware virtualization technology,OEMs can achieve complete isolation between services and applications to ensure that they can still maintain security in the event of system intrusion or data damage.The GPU hardware virtualization solution can support up to eight independent containers to run applications or services at the same time.The auto OEM manufacturer can deploy and remove services and applications at will without affecting the services and applications in other containers running at the same time.In addition to isolation,hardware virtualization technology also supports flexible adjustment of business priority and allocation of GPU performance between multiple car OSs and applications,saving software and hardware costs.
Firmware-based GPU
In addition,in many architectures,hardware graphics events are processed by graphics drivers on the CPU,but the firmware based GPU pioneered by Imagination can process most graphics events internally with ultra-low latency.This method also reduces the host CPU overhead of graphics drivers,and provides a lot of useful additional features,such as collecting all data logs for error analysis when the GPU fails,Connect external cameras or compressed and decompressed devices through GPIO to control power consumption,etc.
“Thanks to such unique core technologies,PowerVR GPU,the flagship IP core product of Imagination,was born for 30 years,providing an industry-leading efficient and flexible solution for 3D,2D graphics and high-performance GPU computing.”Huang Yin further added,”In 2019,Imagination’s GPU architecture has been greatly upgraded,and the A-series AXT GPU has been launched.As the tenth generation core of Imagination,the performance density ratio has increased by 2.5 times compared with the ninth generation.It has a more complete computing logic unit,improves the ability of texture processing,can configure Cache Size,and optimizes the compression algorithm.In 2020,the B-series BXT GPU product has begun to support multi-core technology,and has been optimized in computing power,Micro-architecture has been fine-tuned;In 2021,the C-series CXT GPU launched added a ray tracing(RAC)core to accelerate hardware ray tracing to meet different product requirements.”
From 2019 to 2022,Imagination successively launched powerful GPU IP series
Build a new ecosystem of computing power to meet the challenges of heterogeneous computers
The past 50 years have witnessed the development of algorithms and computing power.The innovation and evolution of computing have focused on scalability,flexibility,efficiency,efficiency and programmability.When the data explosion brings the demand for AI computing,chip companies including SoC Design Company are pursuing solutions with higher performance,lower power consumption and more bandwidth saving.Heterogeneous computing is the only way.This requires the adoption of heterogeneous solutions in the overall design of hardware and SoC,the integration of different types of cores,such as CPU,GPU,NPU,neural network acceleration unit,etc.,and the professional division of labor and more advanced processing of data.
In view of the trend of heterogeneous computing,according to Huang Yin,Imagination will innovate in multiple directions.First of all,GPU will continue to evolve on multi-core architecture and optimize performance and power consumption.Huang Yin pointed out that”Imagination GPU architecture supports 16-bit and 32-bit floating-point operations(FLOAT Operation)and INT 8/16/32 fixed-point operations,while texture processing unit,anti-warping processing,basic filtering,and more complex filtering combined with shaders can be implemented by different operators in GPU.”From the A series,PowerVR GPU uses a new arithmetic logic unit(ALU)architecture and a simpler engine for scheduling,You can use more channels in the same chip area and power budget,fully embrace the design trend of large-scale thread-level parallelism,and simplify compiler work,so as to achieve higher utilization and efficiency.
Imagination’s innovative technology on IP architecture for heterogeneous computing
Imagination also owns the industry-leading neural network accelerator NNA,which adopts the full-hardening neural network acceleration methodology to reach the industry-leading PPA.As the main force and the most efficient part of heterogeneous computing,NNA has more computing power and higher computing power density than CPU and GPU,and carries most of the computing tasks.
Not only that,at the end of 2021,Imagination announced the launch of the new RISC-V CPU core Catapult.As an effective supplement to GPU and NNA,the CPU gives play to its advantages in control flow and flexibility,effectively controls and schedules GPU and NNA for calculation,and also efficiently detects and fills in the gaps on the whole,which is an indispensable part of heterogeneous computing.
Secondly,at the software and system level,Imagination provides a unified,open and standard tool chain for heterogeneous computing,supports standard framework operations,and innovatively optimizes storage and security in order to eliminate data bottlenecks,so that developers can quickly and flexibly deploy their applications on heterogeneous computing architectures.In the whole tool chain,from compilation preparation to the implementation of heterogeneous computing on the device side,Imagination provides developers with rich deployment and optimization functions,including the ability to help developers determine quantitative strategies,meet the requirements of their application performance and accuracy,and leave flexibility to developers to the greatest extent.
In terms of application of heterogeneous computing,Huang Yin said that in the automotive industry,Imagination actively cooperates with chip manufacturers,Tier-1 manufacturers and OEM manufacturers to actively carry out ecological cooperation,provide integrated IP solutions and reshape the interaction with vehicles.At present,Imagination is also the largest GPU IP supplier in the automotive industry,with a market share of about 51%in the cockpit field.
In games,VR/AR and other applications,hardware real-time ray tracing technology was first proposed and implemented by Imagination.Ray tracing can be implemented at different performance and efficiency levels.To clarify this,Imagination has established a Ray Tracing Level System(RTLS),which determines the requirements for ray tracing functions at six levels from Level 0 to Level 5.The Photon architecture adopted by IMG CXT is at Level 4 of RTLS,which is the most advanced hardware ray tracing technology currently available.The application on the mobile end can provide real-time hardware ray tracing and desktop-level quality experience for mobile game players and developers.
Entering the era of real intelligence,digital technology has become a rigid need,which cannot be separated from the support of chips.Whether it is digital consumer applications such as mobile devices,cars and electrical appliances,or infrastructure industries such as data centers and servers,more and more use of advanced IP technology to enable.Huang Yin said:”For more than 30 years since its establishment,Imagination has been committed to providing high-performance IP products and solutions.Imagination’s technological leadership is driven by the market and also leads the market development.In the future,Imagination will also follow the market,improve the scalability,flexibility and efficiency of IP cores,and explore software and data in mobile,automotive,desktop,data center and other diversified markets by creating powerful heterogeneous computing solutions And the value of services.”