91³Ô¹ÏÍø

AI Chip Startups Turn to the Cloud to Compete

Vikram Bhatia

Sep 17, 2024 / 4 min read

The AI chip ecosystem is undergoing a transformation as companies of all sizes compete to address the incredible and fast-growing demand for AI compute. Futurum Research that the processor and accelerator market for AI applications in the data center segment will grow at a 30% CAGR to reach $138B by 2028, up from $38B in 2023.

ai chip startups cloud eda tools

This forecasted growth is astounding, but what¡¯s also notable is the number of AI chip startups in this highly competitive as well as cost- and technology-prohibitive industry, where only the most innovative processors can survive. Talk about barriers to entry! Despite these factors, the AI chip startup ecosystem is well established with an estimated .

How are these companies, with fewer resources and infrastructure compared to today¡¯s established AI chip players, able to innovate and compete? Mirko Prezioso, Co-founder and CEO of Mentium Technologies, a startup specializing in AI co-processors for mission-critical applications in space, robotics, and security, explored this during his session at the 2024 AI Hardware & Edge AI Summit in San Jose, Calif. Prezioso credits his team¡¯s use of cloud services for accelerating the time to tapeout of their most recent design. In fact, a common thread across Mentium and other AI chip startups¡¯ success is their use of cloud-enabled chip development solutions.

ai hardware and edge ai summit

Shown: Mirko Prezioso, Co-Founder and CEO, Mentium, presents at the AI Hardware & Edge Summit 2024 in San Jose, Calif.

Synopsys Cloud Creates a New Playing Field

Synopsys is working with a number of emerging AI chip and IP companies like Mentium, Blue Cheetah Analog Design, Rain.ai, and TetraMem, among others, to advance their development process, from exploration and implementation to manufacturing. The Synopsys Cloud EDA-as-a-Service chip design platform, including the FlexEDA business model along with seamless access to our portfolio of semiconductor IP, has helped customers increase their time to results by 40 percent on average in addition to driving development improvements.

ai chip startups cloud eda tools

For silicon startups, the most precious commodity is time. The faster they can tapeout their prototype design, the sooner they can bring their innovations to market. Synopsys Cloud helps accelerate time to market in three unique ways. 

1: Accelerated Setup Time

Setting up a new EDA environment is a complex undertaking that involves creating the compute and storage environment, uploading EDA software, setting up schedulers and user management, configuring license servers and installing licenses, enabling foundry process design kits (PDKs) and collateral, enabling EDA flows and scripts, and more before the first task can be performed by an engineer. This process can take several weeks, and this is precious time that can be saved by leveraging a Synopsys Cloud SaaS deployment, which provides everything listed above in a preconfigured, secure environment with predefined flows and optimized compute. A startup can be ready to start running EDA jobs and building their chip design within a couple of days of onboarding to the Synopsys Cloud SaaS platform, for example.

effect photonics cloud eda

2: Faster Time-to-Results with Per-Minute Licensing Automation

Synopsys Cloud offers complete license management automation with FlexEDA Pay-per-Use licensing that provides access to unlimited EDA licenses instantly on-demand, by the minute. This capability is transformational for engineers as it can accelerate the entire chip design lifecycle significantly. As engineers pay for actual usage based on the duration of an EDA job for cloud compute resources as well as EDA software licenses, faster time-to-results does not translate into increased costs for the engineer. To illustrate, here¡¯s an example of a SPICE simulation test run with scale up on cloud that accelerates time-to-results by 8X.

SPICE simulation tools

Similar results are seen across the range of Synopsys EDA tools available via the FlexEDA pay-per-use model for a variety of workloads, including verification, sign-off, library characterization, power analysis, and more. For hardware-assisted verification, AI chip startups can leverage Synopsys ZeBu? Cloud for quick-start pre-silicon emulation services without the up-front capital expenditure and internal infrastructure setup and maintenance requirements of on premise capabilities. Startups that have taped-out their designs on Synopsys Cloud have validated these claims for their SoC design projects:

ai chip startups

3: Increased Foundry Confidence

A key step in a startup¡¯s business development is securing support from foundries to manufacture their final designs. This requires validating their business models and demonstrating the feasibility of their designs. Foundries need to validate security and regulatory compliance of the EDA environment before they are able to provide their PDKs and collateral to be uploaded and utilized for design tapeouts. These foundry audits can add weeks-to-months to a startup¡¯s development schedule and can have a significant impact on the startup¡¯s success.

Synopsys maintains strong, trusted relationships with the industry¡¯s leading foundries, including their participation as a Synopsys Cloud partner. In addition, our adherence to SOC 2? Type 2 security and privacy criteria for managing data demonstrates a commitment to safeguarding data in the cloud against unauthorized access, further establishing foundry confidence for a startup.

Thriving in Today¡¯s AI Chip Ecosystem

AI chip startups are boldly challenging ecosystem heavyweights to disrupt architectures and development cycles. Equipped with reliability, flexibility, speed, and security in the cloud, these companies are not only surviving in a highly competitive market but thriving. Learn more about Synopsys Cloud capabilities, Synopsys ZeBu? Cloud for quick-start pre-silicon emulation, and the Synopsys Cloud Startup Program.

Continue Reading