Site icon Converge Digest

Tenstorrent Launches Open Chiplet Atlas Ecosystem

Tenstorrent introduced the Open Chiplet Atlas (OCA) Ecosystem, an open framework designed to standardize and accelerate development across the emerging chiplet market. Unveiled at the company’s San Francisco event, OCA establishes a multi-layered architecture for interoperability across physical, transport, protocol, system, and software layers—paving the way for modular, mix-and-match chiplet integration. More than 50 companies and research institutions have joined the initiative, including LG, Rapidus, Semidynamics, BOS Semiconductors, and the Barcelona Supercomputing Center (BSC).

The OCA Architecture follows a vendor-neutral and ISA-agnostic model intended to lower development costs and reduce design time. It introduces open-source specifications under three main pillars: Architecture, defining interoperability standards across five layers; Harness, a reusable framework that allows developers to focus on value-added logic while ensuring compatibility; and Compliance, a verification and validation program with “Golden Chiplet” testing and industry plugfests. The current draft, OCA Specification v0.7, is available for public review at openchipletatlas.org.

Industry supporters—including foundries, IP providers, and system designers—view the OCA as a path toward greater silicon modularity and supply chain diversity. Tenstorrent positions the initiative as a foundational layer for heterogeneous computing, from AI accelerators to automotive and data center SoCs. Academic backers such as Oxford University and the University of Tokyo are contributing research toward scalability, reliability, and performance optimization within the open chiplet model.

“The future of silicon is heterogeneous and composable,” said Wei-han Lien, Chief Architect at Tenstorrent. “The OCA Ecosystem establishes the foundation of trust and interoperability needed to unlock the full potential of multi-vendor chiplet-based SoC products. This is not just a new standard—it’s the beginning of a collaborative community that will drive the next generation of system innovation.”

🌐 Analysis:

The Open Chiplet Atlas initiative reinforces growing industry momentum toward disaggregated silicon architectures, complementing parallel efforts such as OCP’s Open Chiplet Economy and the Universal Chiplet Interconnect Express (UCIe) consortium. For Tenstorrent, which already champions open RISC-V cores and modular AI compute, OCA represents a strategic extension into chiplet-level interoperability. As foundries like Rapidus and ITRI support the framework, the ecosystem could accelerate standardized chiplet manufacturing and packaging, a crucial enabler for the next wave of composable AI systems and custom SoCs.

Tenstorrent, headquartered in Toronto and founded in 2016, develops next-generation computing hardware and software platforms for artificial intelligence workloads. The company’s mission is to deliver scalable, high-performance AI systems through open and flexible architectures that span data center, edge, and embedded applications. Its core technology is built around the proprietary Tensix processor architecture—combining RISC-V control, tensor compute units, and a high-bandwidth network-on-chip fabric—implemented in chips such as Grayskull, Wormhole, and Blackhole. Tenstorrent’s software stack, including its TT-Forge compiler and TT-Metalium runtime, enables developers to deploy machine learning models efficiently across heterogeneous environments.

Led by CEO and renowned chip architect Jim Keller, the company has attracted more than $1 billion in funding from strategic investors including major technology and automotive groups. Tenstorrent’s roadmap emphasizes modular, chiplet-based designs and open standards that foster interoperability and lower development costs, exemplified by its recent launch of the Open Chiplet Atlas (OCA) Ecosystem. With deep RISC-V integration, a growing patent portfolio, and partnerships spanning foundries and system integrators, Tenstorrent is positioning itself as a leading alternative to GPU-centric AI acceleration, targeting both hyperscale and embedded markets.

🌐 We’re tracking the latest developments in networking silicon. Follow our ongoing coverage at: https://convergedigest.com/category/semiconductors/

🌐 We’re launching the “Data Center Networking for AI” series on NextGenInfra.io and inviting companies building real solutions—silicon, optics, fabrics, switches, software, orchestration—to share their views on video and in our expert report. To get involved, send a note to jcarroll@convergedigest.com or info@nextgeninfra.io.

Exit mobile version