Athos Scraps Multi-Vendor Roadmap, Plans Chiplet Tape-Out

2026-04-30 · Source: Big Data & AI News - EE Times · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Advanced, medium

Summary

Athos Silicon, a functional safety chiplet startup spun out of Mercedes-Benz in October 2025, has revised its technology roadmap to focus on a single, in-house-designed SoC chiplet. This strategic shift, driven by Arm's $240 million acquisition of DreamBig's chiplet hub, replaces an earlier plan for a multi-chiplet SoC using third-party compute and NPU dies. The new design integrates third-party CPU, GPU, and NPU IP into a single Athos-designed chiplet, aiming for aggressive pricing and enhanced control over functional safety aspects like binning and power management. Athos's patented voting system allows identical compute chiplets to monitor neighbors for issues, providing full error disambiguation, unlike traditional lockstep methods. The company also incorporates Mercedes-Benz's trajectory choice algorithm, which uses laser and radar data to rank trajectories by safety, aiming for a system statistically one million times safer than ISO 26262 minimums. First samples are expected in Q1 2027, targeting autonomous driving and drones.

Key takeaway

For AI Architects designing safety-critical autonomous systems, Athos Silicon's shift to a single, in-house chiplet design highlights the importance of cost control and deep integration for functional safety. You should evaluate whether a custom, consolidated chiplet approach could offer better pricing competitiveness and superior error disambiguation compared to multi-vendor, multi-chiplet solutions, especially given stringent automotive lifetime requirements and liability concerns. Consider the benefits of a voting system for fault tolerance over traditional lockstep methods.

Key insights

In-house chiplet design offers cost control and superior functional safety for autonomous systems.

Principles

• Cost-effectiveness drives single chiplet designs.
• Full design control enhances functional safety.
• Voting systems improve error disambiguation.

Method

Athos's design uses identical SoC chiplets with a patented voting system to monitor neighbors for hardware/software issues, enabling error disambiguation and dynamic recovery. It also employs a non-ML trajectory choice algorithm for enhanced safety.

In practice

• Implement redundant power rails for safety-critical systems.
• Utilize idle chiplets for built-in self-tests (MBIST/LBIST).
• Integrate non-ML algorithms for safety-critical decision-making.

Topics

• Athos Silicon
• Functional Safety Chiplets
• Autonomous Driving Systems
• Chiplet Design Strategy
• Error Disambiguation Voting

Best for: Investor, Entrepreneur, AI Hardware Engineer, AI Architect, Robotics Engineer

Related on AIssential

• Chiplets, Ecosystems, and Europe’s Post-Fab Semiconductor Strategy — Europe shifts its semiconductor strategy from fab-centric to chiplet-driven ecosystem leadership, responding to AI's demands.
• Claude Code, Codex and Agentic Coding #8 — AI coding agents are rapidly advancing but require careful management to mitigate risks from autonomous actions.
• Distilling Safe LLM Systems via Soft Prompts for On Device Settings — Soft prompt distillation effectively transfers guard model safety to LLMs for resource-constrained on-device deployment.
• The Alignment Flywheel: A Governance-Centric Hybrid MAS for Architecture-Agnostic Safety — Decouple autonomous decision-making from safety governance using a hybrid MAS for auditable, patchable oversight.
• Google plans nearly two million new AI chips as it turns to Marvell for custom designs — Google is developing custom AI chips with Marvell to enhance inference capabilities and diversify its supply chain.
• TDPNavigator-Placer: Thermal- and Wirelength-Aware Chiplet Placement in 2.5D Systems Through Multi-Agent Reinforcement Learning — Multi-agent reinforcement learning can resolve conflicting objectives in 2.5D chiplet placement.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Big Data & AI News - EE Times.