Samsung and VeriSilicon Enable Blaize to Meet Aggressive Time-to-Market Goals for New AI Edge Processor
Collaborative effort completed on schedule despite emergence of COVID-19 pandemic immediately before tape-out of Blaize Graph Streaming Processor device
SAN JOSE, CA – Oct 28, 2020 – Samsung Electronics Co., Ltd., a world leader in advanced semiconductor technology, in collaboration with design services provider, VeriSilicon, has successfully supported the on-time market launch of AI Edge computing startup Blaize’s hardware platform, despite the COVID-19 pandemic.
Leveraging Samsung Foundry’s 14nm FinFET process technology and VeriSilicon’s chip design expertise and IP, Blaize is now offering its AI Edge computing Pathfinder and Xplorer platforms equipped with the Blaize Graph Streaming Processor (GSP), bringing a new class of powerful AI compute capabilities to edge applications such as industrial, smart city, automotive sensor fusion, last mile delivery and retail.
Blaize, based in El Dorado Hills, Calif., is addressing a fundamental issue in Edge AI application deployment: existing solutions are either unable to handle the required computational load, or too costly and difficult to implement in products. At the heart of Blaize’s solution is a 100% programmable hardware platform that offers an unprecedented combination of low-latency AI processing, low energy demand, multi-sensor AI analytics, and an accompanying suite of software tools that greatly simplify and accelerate the AI development process.
“We built our architecture for Edge AI, rather than adapting it from designs developed for larger settings like data centers and Cloud applications,” explains Santiago Fernandez-Gomez, Blaize vice president of platform engineering. “Our customers are asking for solutions that are a fraction of the size, use much less power, and can be tailored to their specific market requirements.”
Blaize’s platform strategy implementation faced a critical challenge: quickly bringing a robust product to market. “For both technical and business reasons, our number one goal was to get into production and have products to sell, so we needed working GSP chips as soon as possible. That made a mature process node a top priority,” notes Fernandez-Gomez.
At the outset of Blaize’s engagement, Samsung Foundry assessed initial plans and goals for the GSP and recommended involving VeriSilicon, a one-stop source for custom silicon solutions and semiconductor IP, with an extensive track record including other AI-related projects.
“The Blaize GSP project is a perfect example of why Samsung Foundry has prioritized development of the Samsung Advanced Foundry Ecosystem (SAFE) partnership program, as we pursue our goal of delivering flawless execution,” states Hong Hao, senior vice president of Foundry Business at Samsung Electronics. “Providing well-integrated access to VeriSilicon’s design and IP expertise allows a young and innovative company like Blaize to leverage the full breadth and value of our leading process technologies in the most efficient way.”
“It is a compelling chip design, and the Blaize team has a very good technical background and competence in collaborating across multiple sites,” recalls Dr. Mahadev Kolluru, VeriSilicon’s vice
president for North America platform solutions sales. An early order of business was selecting the optimal manufacturing process node; after evaluation of many options, Samsung Foundry’s 14nm FinFET process emerged as the ideal choice for this product.
“When we considered the target applications for the initial chip, and the fact that much of the required IP at the time was not yet available in more advanced nodes, the mature node made sense,” says Dr.
Kolluru. “Cost is always a concern for everyone. And given our experience working with Samsung on this process over the years, we were confident it was the best path to first-time silicon success.”
“In those initial few months, the Samsung Foundry team worked hand-in-hand with VeriSilicon, going through every possible design method that would give us the combination of performance and power consumption we required,” recalls Fernandez-Gomez. “The Samsung foundry brought solutions to the table beyond the standard flows that allowed us to fully meet out targets.”
Another key factor was effective supply chain planning. Despite the Covid-19 pandemic and global materials supply shortage, VeriSilicon was able to reserve production capacity not only with Samsung foundry but also with packaging, assembly, and test contractors.
The process rolled forward smoothly through 2019 and into 2020, with teams across locations working together in seamless fashion to address design challenges like system timing and integration of complex machine-learning IP blocks with the graphic-processing elements of the chip. VeriSilicon also provided a verification platform that allowed Blaize to begin developing drivers for the hardware platform. All systems were ready for scheduled tape-out in February when the global COVID-19 forced the lockdown of companies worldwide. VeriSilicon had to close offices 2 weeks before tape-out, but quickly adapted to working remotely along with the Blaize teams who were also transitioning to working remote, to deliver on the original target date.
Even better, when first silicon arrived, the samples were good enough to go directly into production. “In my experience,” adds Fernandez-Gomez, “it can often take six to nine months to go from first silicon to customer samples, and 12 to 18 months to a full production run. VeriSilicon delivered customer samples in two months and the silicon was pretty much perfect, and we will be in real production after six months.”
Samsung’s Hao calls the project “a success story that shows how a collaborative team effort can solve tough problems and meet ambitious goals. It is tremendously satisfying to work alongside people of the caliber found at Blaize and VeriSilicon, and to see so much knowledge and capability coming together in service of these exciting new Edge AI applications.”
|
Verisilicon, Inc. Hot IP
High performance, energy efficient 3-issue, quad-mac superscalar DSP core for H ...
Neural Network Processor for Intelligent Vision, Voice, Natural Language Process ...
High performance, energy efficient 4-issue, 16-mac vector DSP core targeted for ...
Highly efficient 2-issue, dual-mac superscalar DSP core for Voice, Audio applica ...
Related News
- SiFive and ArchiTek Enable Secure, Private, Flexible Edge AI Computing With AiOnIc Processor
- Verisilicon High-Performance and High-Quality AI Video Processor Powers Leading Datacenters
- DSP Group Unveils DBM10 Low-Power Edge AI/ML SoC with Dedicated Neural Network Inference Processor
- Blaize Delivers First Open and Code-free AI Software Platform Spanning the Entire Edge AI Application Lifecycle
- Lattice Propel Accelerates Time-to-Market for Embedded Processor-based Designs on Latest Nexus Platform FPGAs
Breaking News
- Expedera Introduces Its Origin Neural Engine IP With Unrivaled Energy-Efficiency and Performance
- PLDA Introduces a Complete Line of PCIe IP for USB4, Enabling PCIe Support in USB4 Hubs, Hosts and Devices
- DSP Concepts Enables Audio Weaver for the Cadence Tensilica HiFi 5 DSP
- Logic Design Solutions Introduces the first NVMe HOST IP on POLARFIRE SoC FPGA
- TSMC Boosts Capital Budget Again, to $30B
Most Popular
- Synopsys To Expand DesignWare Ethernet IP Portfolio with Acquisition of MorethanIP
- Siemens expands industry-leading IC verification portfolio through acquisition of OneSpin Solutions
- Arm and NVIDIA: Fueling Innovation for the Next Era of Compute
- OpenFive Tapes Out SoC for Advanced HPC/AI Solutions on TSMC 5nm Technology
- Avery Design Debuts CXL 2.0 System-level VIP Simulation Solution
![]() |
E-mail This Article | ![]() |
![]() |
Printer-Friendly Page |