In recent years, with the rapid development of emerging technologies and industries such as AI, 5G, Internet of Things, and cloud computing, the amount of data is exploding at an immeasurable rate. According to IDC**, the total amount of data in the world will grow nearly 20-fold from 2015 to 2025 to 175 zettabytes.
In the era of science and technology, massive data has generated a huge demand for computing power, which in turn has driven high-performance and high-computing chips to become the driving force and new trend of industry development.
While the market demand for high-performance chips is increasing, it also further accelerates the iteration cycle of chips, stimulating chip design companies to continuously improve chip technology content and shorten delivery time to enhance core competitiveness. On the other hand, with the iteration of products and the more and more advanced manufacturing processes, the demand for IT resources and the growing development cycle and cost brought by chip design have gradually become the pain points and challenges in front of chip design companies.
Especially in recent years, the number of small and medium-sized chip design companies in China has been increasing. According to statistics from the China Semiconductor Industry Association, the number of chip design companies in China has increased from 736 in 2015 to 3,451 in 2023, of which about 84% are small and micro enterprises with less than 100 employees.
Limited by their own scale, strength and technical accumulation, small and micro enterprises cannot fully meet their own rigid needs such as manpower, computing power, and software in the chip development and design process, which is easy to cause practical pain points such as low resource utilization, high R&D costs, and slow transformation of technological achievements.
A "roadblock" on the road to chip design
It is not difficult to find that modern chip design, especially high-end chip designs such as CPU, GPU, NPU, etc., has several times the workload of computing, verification, and ** than in the past.
For small and medium-sized chip design enterprises, there will be a shortage of resources when the peak demand for verification comes, and spending a lot of money to purchase servers for physical verification does not meet the requirements of the company's development stage.
Because the demand for computing power at different stages of project development fluctuates. Generally, the chip design cycle is generally more than 12 months, including product definition, front-end design, IP verification, SOC verification, synthesis, place-and-route and other stages, and the requirements for computing power are different at different stages.
Diagram of the relationship between chip development cycle and IT resource allocation.
In this case, it is difficult for chip design companies to accurately balance resource demand and supply, and if there is more waste, less they have to repeatedly purchase, build and maintain, even if they are prepared according to the intermediate value, when the number of tasks at a certain point in time surges, there will be a mismatch of resources.
Even for large chip design companies, there is usually a competition for resources between different project teams, and the company's resources are always limited, and it is difficult to be flexible in resource allocation, and there is often a mismatch between "queuing" and "idle" at the same time. This mismatch will lead to low resource utilization and affect the progress of R&D.
It can be seen that for the majority of chip design companies, the balance between computing resource planning and cost is an art.
On the other hand, the chip design environment is complex, and it is difficult for enterprises to build. Because of the general lack of professionals in the chip design industry, small and medium-sized companies do not have a dedicated IT CAD team for the deployment and maintenance of the R&D environment, so if they want to completely build a local R&D and design environment, the cycle is long and complex, and the personnel cannot keep up, which requires huge labor costs.
On the whole, "lack of resources, lack of manpower, lack of tools" and so on have become common problems faced by chip design companies
Resources are urgently needed:Enterprises need a lot of computing power for verification before tape-out, and it is difficult for local resources to meet them.
Lack of professionals:The cluster environment is complex, and cluster O&M is difficult.
Professional management tools are lacking:Lack of easy-to-use cluster management and scheduling software makes it difficult to manage the cluster environment.
Do you have enough server resources to deploy it quickly? How to respond to the soaring demand for computing power? How to effectively balance the company's R&D costs? What is the solution to the dilemma of a shortage of IT CAD engineers and a complex chip design environment that is difficult to build?
A series of problems have undoubtedly become a "roadblock" in front of chip design companies.
Under the challenges, how can chip design companies break the situation?
Chip design opens the era of "going to the cloud".
In this way, chip design companies should not only consider the elastic demand of project computing power, but also avoid unnecessary time consumption.
Bringing chip designs to the cloud seems to be the best solution at the moment.
After the chip design is migrated to the cloud, the company can purchase and expand from cloud providers on demand, which can flexibly and quickly meet the fluctuating demand for computing power. After the peak demand for computing power has passed, cloud resources can be quickly released, thus saving related costs.
This resiliency provides developers with flexibility, agility, and scalability, improving resource utilization. Especially under the trend of continuous development of chip manufacturing process and higher and higher degree of chip integration, a large number of computing, verification, and advanced work all need strong IT resources as support, and the manpower and material resources invested in building IT resources and equipment by enterprises alone will be very large.
Therefore, migrating chip design to the cloud will become the only way for chip design companies to "reduce costs and increase efficiency".
To put it simply, the essence of chip cloud migration is to exchange computing power for time, use computing power to save manpower, balance cost investment through cloud elastic computing power, and liberate productivity.
Compared with the past mode of deploying local computing power by chip design companies, the cloud platform can flexibly use resources and open them at any time, which can achieve smooth fluctuations with the computing power demand curve and high resource utilization to meet various business emergency needs.
It is foreseeable that the cloud is becoming a trend in the chip design industry, and the era of IC design in the cloud and flexible adjustment to meet the demand for computing resources has arrived.
In this regard, cloud providers are also actively deploying.
As early as 2017, several EDA vendors such as Synopsys and Cadence in the United States began to put forward the concept of chip design to the cloud, and cloud computing companies such as AWS and Microsoft provided chip cloud solutions.
For example, Synopsys and TSMC jointly deployed a cloud-based design and chip manufacturing platform, helping TSMC become the first foundry to implement cloud design. Synopsys also partnered with TSMC to implement a new cloud-based timing signoff process on the Microsoft Azure platform to accelerate system-on-chip development efficiency.
In Europe and the United States, it has been a market consensus to migrate chip design to the cloud, and mainstream chip design manufacturers, foundry, and EDA companies at home and abroad have switched to cloud design**. Compared with foreign countries, the beginning of domestic chip design to the cloud began around 2019, and many cloud platforms, including Alibaba Cloud, Tencent Cloud, Unisplendour Cloud, and Huawei Cloud, began to explore high-performance computing scenarios.
In addition to solving the pain points of the elastic computing power demand of chip design, the public cloud also has a rich and comprehensive ecological cooperation system, sales chain collaboration, and provides more comprehensive services for chip companies.
However, there is no absolute standard for how to choose a cloud solution in the industry, and there are too many cases to learn from.
In addition to the pay-as-you-go public cloud method, there is also a private cloud method, that is, to build a chip design cloud locally to meet its own chip design needs. Because as some chip design companies become larger and larger, chip design needs some virtualization needs in addition to computing power elasticity, so it has gradually become a trend to build a chip design private cloud to use and improve the fixed resource pool.
If a large company buys all the public cloud, the cost will be relatively high, and it is more inclined to build a fixed resource pool locally, use the resources of the private cloud to meet daily use, and then involve the elastic computing power part and combine with the public cloud to form a hybrid cloud model.
In addition, some companies have deployed some hardware equipment, and going to the cloud does not mean moving all the local equipment to the cloud, but also through partial cloud, that is, local equipment and resources are reserved as a standing computing power, supplemented by the elastic supplement of computing resources on the cloud, and the two synergy may be a better cost-effective solution.
It is understood that at present, AWS, Azure, Alibaba Cloud and other leading cloud providers mainly provide public cloud, and there is less involvement in the deployment of chip design private cloud, and more enterprises such as Unisplendour Cloud, Speedstone Technology, and Ivy Open Source Technology (HPC Insight) are in the layout, both public cloud and private cloud.
It is worth noting that, compared with Speedstone Technology and Ivy Open Source Technology, which rely on third-party cloud platforms and underlying resources, the public cloud of Unisplendour Chip Cloud relies on its own Unisplendour cloud platform, and works with the companies of Unisplendour Group to do private cloud scenarios in terms of the underlying resources of private cloud, creating a complete chip design environment for chip design enterprises in an all-round way, starting from key links such as computing power, CAD tools, cluster scheduling and management, to achieve the effect of reducing costs and increasing efficiency.
In short, the cloud on the chip design platform is no longer a difficult choice. Public, private, and hybrid clouds are all effective solutions to the challenges of high-performance chip design, but there are some differences between them, and users can choose the path and mode that suits their needs.
It's hard to build a chip design environment!
It can be seen that the cloud of chip design can help enterprises achieve a value balance in computing power demand, efficiency and cost.
But back to the issue of helping enterprises shorten the chip design cycle, only providing computing power support in the cloud is only a basic service. Due to the lack of full-time IT CAD engineers and other talents, it is difficult for some enterprises to quickly build a chip design environment even if they get cloud computing resources.
As mentioned at the beginning of the article, the chip design environment is complex, lacks professionals and management tools, and lacks easy-to-use cluster management and scheduling software, which makes it difficult to manage the chip design cluster environment and operate and maintain.
In this regard, building a professional design environment has become a new challenge and demand for chip design enterprises.
From the perspective of the current situation of the industry, industry users are most familiar with IBM's cluster core software LSF, which is responsible for the management of computing resources and the scheduling of batch processing jobs, which provides users with a unified cluster resource access interface, allowing users to transparently access the entire cluster resources, and at the same time providing rich functions and customizable policies.
At present, LSF is basically an indispensable basic software in high-performance computing environments.
In the field of chip design, LSF scheduling tools usually cooperate with cloud providers such as Azure and AWS to help LSF schedule cloud resources more efficiently.
Domestically, there are also many companies trying here. For example:
Speedstone Technology has launched an IC design and R&D cloud platform to manage the complex R&D environment of enterprises locally and in the cloud in a unified and collaborative manner.
Ivy Technology's HPC Insights monitors and intelligently analyzes HPC clusters, helping customers fully control cluster resources and optimize cluster resource allocation.
Moore Elite focuses on the "chip design to the cloud" business to provide users with an efficient and easy-to-use R&D environment.
In addition, we also see that some companies are trying to build a platform, aiming to use AI technology to help chip design companies optimize IC design, and strive to design chips with the best performance in the optimal mode.
It is worth noting that in the field of domestic chip design on the cloud, Unisplendour Cloud is the first practitioner and promoter of the industry.
It is understood that as one of the earliest industry practitioners to promote the chip design cloud in China, as early as around 2020, Unisplendour Cloud began to build and deploy the chip design cloud in advance.
Geng Jiashen, chief architect of Unisplendour Chip Cloud, pointed out that the development background and opportunities of Unisplendour Chip Cloud are mainly due to the internal needs of the group, and some chip companies under the group want to use some public cloud resources to meet the demand for elastic resources.
In August 2020, Unisplendour launched Chip Cloud 10 version, officially promoted to the outside world; After less than half a year, the Unisplendour chip cloud was upgraded again and released 2Version 0.
It is understood that the purple chip cloud 10 focuses more on the provision of computing power and design environment to meet the pain points and needs related to computing power and design environment. And the upgraded purple chip cloud 20 services are more comprehensive - by providing a professional privatized chip design environment and flexible public cloud resources, it provides a complete hybrid cloud solution for chip design enterprises.
It can be understood that the original purpose of Unisplendour Chip Cloud was mainly to serve chip design companies within the group. However, in this process, we have seen the common needs of some chip design enterprises, especially in recent years, a large number of small and medium-sized chip design enterprises have emerged, which has derived a large demand for computing resources and the construction of the entire chip design environment.
Therefore, through the accumulation of experience and technology in the process of serving internal enterprises, the Unisplendour chip cloud business has formed a standardized chip solution, which is then promoted and used to customers in the industry.
In addition to computing power, it also helps enterprises sort out the business status, build a chip design environment, and be responsible for technical support and operation and maintenance services in the later stage by virtue of a localized professional CAD IT team and rich practical experience in chip design on the cloud.
Among them, the two key products in the chip design environment are Zixin scheduling software and CAD management platform.
Facing chip design scenarios, the Purple Core Scheduler can quickly build and manage heterogeneous high-performance computing clusters in data centers and clouds, providing users with elastic resources and a unified resource scheduling management platform for business needs, and improving resource utilization.
In the face of the current situation of scheduling software relying on overseas products, Zixin Scheduler aims to solve the problem of industry bottlenecks and realize the localization of scheduling software. Geng Jiashen said, "Based on the addition of IBM's LSF, the purple core scheduler has further developed new functions and features such as multi-cluster scheduling and automatic parameter completion and replacement in the process of submitting job parameters, and optimized and improved the performance bottleneck of the open source scheduler in large-scale scheduling. ”
In short, in the face of the traditional advantages of LSF, an old competitor in the industry, Zixin Scheduler chose to find a new way, combining the industry's latest container scheduling and AI and other new technologies to make some improvements and optimizations to achieve differentiated competition.
Some industry insiders said that many customers who have been contacted in the past have used imported schedulers, but when switching to domestic tools, it is very troublesome to adapt and convert the original R&D process and schedule. Geng Jiashen said that the self-developed domestic purple core scheduler can be compatible with international schedulers from the bottom, solving the problem of "stuck neck" while retaining the developer's past usage habits. Combined with CAD capabilities and experience, the Unisplendour chip cloud platform can effectively improve the overall linkage efficiency of the upper, middle and lower layers, help customers reduce costs and increase efficiency of chip design, and can jointly adapt to landing scenarios.
This CAD capability refers to the CAD management platform of Unisplendour Chip Cloud, which is a one-stop management platform for IT CAD operation and maintenance management in the IC design industry, which horizontally supports unified scheduling of multiple scenarios, and vertically supports the operation monitoring of computing power, storage, and jobs, as well as the intelligent scheduling of clusters and tasks, which can provide users with fast, stable, highly reliable, efficient and low-cost cluster management services, create a professional chip design environment, and improve the efficiency of chip research and development.
It is reported that the CAD management platform of Unisplendour Chip Cloud can manage both its own Zixin scheduler and LSF scheduler. Because many customers have already purchased LSF software, the update speed of LSF's management software is slow, and there will be a problem of missing features based on the old architecture design, and it will not be updated quickly and iteratively according to some user needs.
In contrast, the Unisplendour Cloud CAD management platform can quickly respond to customer needs and update and iterate functions. Therefore, in the current situation of the industry where the market share of LSF scheduler is still generally large, some visual management and monitoring of LSF scheduler can be supported through the CAD management platform, and the localization penetration of the core software of the chip design environment can be gradually realized.
On the whole, as a well-structured and out-of-the-box cloud chip design platform, Tsinghua Unigroup Chip Cloud provides chip design enterprises with a powerful, flexible and secure computing cluster and storage resource pool, equipped with a team of senior IT CAD experts, combined with the best practices of chip design enterprises within Tsinghua Unigroup Group, to help chip design enterprises build a dedicated and private cloud design environment, and provide one-stop design services + cloud service packaging turnkey solution delivery.
Compared with other platforms, Unisplendour Chip Cloud has multiple advantages such as high performance, high security, hybrid deployment, best practices, and ecological operation.
Customer effect is the first persuasion
For the chip industry, there is still a lot of room for imagination in designing to the cloud. However, the premise of achieving more landing applications is that the cloud platform must not only ensure high efficiency, but also gain enough trust from chip design companies.
The Unisplendour chip cloud has an advantage here. According to Geng Jiashen, the Unisplendour chip cloud was launched based on the overall strategic layout of Unisplendour Group from core to cloud, and before it was introduced to the market, it has been fully verified by the chip design enterprises within Unisplendour Group, forming the best practice of chip design on the cloud.
In actual deployment, the most concerned question for chip design companies is whether they can improve resource utilization, and what is their own resource utilization rate? Only after understanding these can we make better adjustments and optimizations.
In response to this problem, the Unisplendour chip cloud platform has achieved good results in helping customers reduce costs and increase efficiency
At present, many chip design companies have set up R&D centers in many places, and the Unisplendour cloud chip team has established a centralized and exclusive cloud chip resource pool for its scattered resources, and then deployed the CAD management platform and Zixin scheduler at the same time for analysis and optimization, increasing the customer's CPU computing utilization rate from 50% to 80%, helping users improve the overall efficiency.
In addition, the visual management software in the chip cloud platform can also help customers select hardware equipment, and help the reasonable procurement of hardware resources such as servers and memory by analyzing the ratio of hardware resources, so as to achieve cost savings.
It can be simply understood that for cross-regional and even multinational chip design companies, which involve the problems of low R&D R&D efficiency and unreasonable hardware resource procurement in multiple places, the Unisplendour cloud chip platform can realize the unified management of cloud and local computing power through clusters, effectively solve the problem of multi-place collaboration, and improve resource utilization; At the same time, optimize the customer's hardware ratio and reduce unnecessary cost waste.
With the blessing of many advantages, Unisplendour Chip Cloud has boosted the confidence of many chip design companies.
In addition to the huge computing power demand and IT CAD design environment, chip design also has EDA tools, IP, PDK, etc., which are the parts with the highest cost and the largest investment for chip design enterprises. For the vast majority of small and micro chip design enterprises, they do not have much bargaining power, and with the help of the existing ecological partners of the Unisplendour chip cloud platform, these can be better supported.
Geng Jiashen said that Unisplendour Chip Cloud has cooperated with a number of EDA and IP partners to provide design tools and IP services. Through in-depth cooperation with the upstream and downstream industry chains of the chip industry and tool manufacturers, we will jointly build full-stack services to achieve a win-win situation.
Through the chip cloud platform, empower small and medium-sized chip design enterprises and empower chip design speed" - such an industrial grand plan and development vision are slowly opening up in the direction of helping the development of China's chip industry under the comprehensive layout of Ziguang chip cloud.
Conclusion
Some people in the industry complained, "When you do chip design in a large company, you don't need to worry about IT, you just need to complain about IT; When you come to a small company, you will find that IT is full of problems - computing power, storage, cost, tools, security, environment, etc., all need to be considered. ”
The purpose of "chip design to the cloud" is to liberate chip design companies from these things and devote themselves to their core business.
Looking at the industry pattern, there are many masters in the public cloud field such as AWS, Azure, and Alibaba Cloud, and Speedstone Technology and Ivy Open Source Technology are eager to try private cloud business, and IBM and LSF are unique in the field of cluster core software.
With the output of Ziguang Group's core cloud capabilities, the industry all-rounder that can deploy both public cloud and private cloud - Ziguang Chip Cloud is committed to helping chip companies build a professional design environment and cope with the challenges of computing power demand through the support of Ziguang Cloud + chip special management platform (CAD management platform + Zixin scheduler), greatly realizing resource sharing, technology sharing, and platform sharing, and accelerating the new upgrading of China's semiconductor industry.
Unisplendour chip cloud business inquiry
Contact: Ziguang Cloud**Customer Service (**Same number as WeChat).