The importance of the software stack for future chips
Hardware/software full stack strategic advantage
A new hardware baseline intended to strengthen European strategic autonomy in the semiconductor market is fundamental (especially in sectors such as edge AI) but represents only the first enabling step of a more solid and forward-looking strategy aimed at this market consolidation and strengthening. The sole availability of European hardware platforms does not ensure the concrete possibility of reaching 20% of the global semiconductor market; only the availability of full stacks supporting the new hardware baseline can allow the market to really take off and concretely support vertical application. This is a lesson we learned 15 years ago (“bare metal doesn’t sell”), which is even more relevant today with the proliferation of hardware platforms in terms of both providers and technology domains (15 years ago, we were mainly focusing on PC or PC-equivalent hardware).
In a recent interview with Intel's CEO Pat Gelsinger, he stated that “if you want your hardware to succeed, you have to put software first.” He is convinced that it is today necessary to work with a broader ecosystem of independent software vendors and to work more closely than before. But, on the other hand, your own software could bring new revenue streams: "One of the things that I've learned in my 11-year 'vacation' [at VMware and EMC] is delivering silicon that isn't supported by software is a bug," he said. "We have to deliver the software capabilities, and then we have to empower it, accelerate it, make it more secure with hardware underneath it. And to me, this is the big bit flip that I need to drive at Intel."
Many other brands are following the same strategy: Nvidia, for example, began to aggressively promote its CUDA platform, whereas other companies relied on various open or proprietary standards like OpenCL, Vulkan, Metal or OpenAI to speed up performance-hungry workloads with proprietary hardware.
The availability of the full stack is a key enabler of the ECS value chain up to the applications (both in Europe and globally) and serves as a critical element in enabling the multiplying factor in revenue streams across the entire value chain, as highlighted by the Advancy report. And this is particularly true for mega-trendsetters such as AI and HPC (embedded HPC, in the case of our community).
The issue of existing and open stacks
Moreover, to achieve a real and solid strategic autonomy, Europe cannot depend on third-party stacks which hide the customer lock-in issue. Companies like Apple and Nvidia created their own software ecosystems, but their intention is to consolidate/increase the customer base, strengthen the customer lock-in, impose their ‘standard’, etc.
It is also important to support existing and future open stacks. This is another way to activate the multiplying factor that increases revenues across the entire ECS value chain, from semiconductors up to the application.
Unfortunately, part of the new European hardware baseline will introduce completely new approaches to computing that cannot be supported by existing computing models and related stacks, therefore requiring completely new stacks.
Stack HW/SW timeline dependency and importance for strategic autonomy
Timewise, the design and development of the stack must proceed in parallel with the design and development of the European hardware baseline. Designing and developing a stack two to three years after the availability of the hardware baseline will introduce a critical delay that will prevent the practical usage of the hardware baseline itself at the crucial moment when it would be ready to be launched on the market. The entire hardware and software stack must be designed and developed as a single entity from the beginning.
Again, this is a lesson already learned, which is particularly relevant today because the ECS product lifecycle has been significantly reduced (from five to seven years in the PC era to two or three years in the IoT, embedded computing and AI era).
What is covered by the HW/SW stack?
The concept of a HW/SW stack that we are considering is covered in full in the ECS-SRIA 2021 and 2022 (structurally covered!) and includes, for example:
● the hardware platform
● the firmware layer
● the operating system support
● the virtualisation support
● the support for distributed computing
● development methods, tools and libraries
● standard development kits (SDKs)
Engineering support and stack automation engineering
Generally speaking, the stack must include engineering support, which is crucial to providing all of the tools to effectively, efficiently and securely exploit the resources, capabilities and functionalities of the new hardware baseline. And this is particularly true for domains such as neuromorphic computing, where engineering tools are completely missing and we cannot rely on the existing expertise and background.
Moreover, the software stack should be developed automatically: this is a crucial aspect to follow the pace of current technology trends and to support the increasing number of hardware platforms. It was indeed already difficult to build a stack in the PC era, and this difficulty has certainly increased in the last decade due to the higher complexity and heterogeneity of ECS. This has a negative impact on stack availability, development costs, quality of developed applications, etc. Also, the ubiquitous presence of software in products, services, operations and management calls for the next industry transformation: the automation of software development! For our industry, acquiring a leading position in the automation of labour-intensive software development represents a crucial element to a competitive, sustainable and autonomous Europe.
 https://www.crn.com/news/components-peripherals/intel-ceo-pat-gelsinger-the-silicon-man-with-the-software-plan  Embedded Intelligence: Trends and Challenges. A Study by Advancy, Commissioned By Artemis Industry Association. https://www.inside-association.eu/publications