Analyse de marché
Intel 18A is not a new x86 processor architecture, but rather a 2nm-class manufacturing process node where Intel will fabricate future CPUs, SoCs, and potentially chips for external clients. This process technology is highly anticipated, as it is central to Intel’s strategy to overcome nearly five years of problems, aiming to mark a significant « before and after » point for the company. The full-scale rollout, initially expected in the second half of 2025, is now projected toward 2026. Marketing claims suggest Intel 18A offers up to a 15% improvement in performance per watt and over 30% higher transistor density compared to previous nodes. PowerVia, a key feature, supposedly improves standard cell utilization by 5–10% and boosts performance by up to 4% at the same power level by optimizing power delivery. Early integration, such as the next-gen Arc B390 graphics built on 18A, reportedly achieves 70% higher frame rates in gaming than AMD counterparts and 50% better AI inference performance. Furthermore, the new process and architecture optimizations are claimed to enable up to 27 hours of battery life in devices. Implementation plans include the launch of the first consumer Core Ultra Series 3 processors (Panther Lake) by the end of 2025, with significant volume ramp-up expected in the second quarter of 2026. Server processors (Clearwater Forest) optimized for cloud computing are scheduled for the second half of 2026, followed by the adaptation of the process for external customers (Intel 18A-P) in 2027. The process relies on two critical architectural innovations. First is the **RibbonFET** transistor structure, replacing the older FinFET design below 4nm. RibbonFET surrounds the current-carrying channel with a gate on all sides (in a ribbon-like shape), providing superior control over electrical current, crucial for reducing leakage, improving efficiency, and enabling scalability at high current densities where FinFET struggles. Second is **PowerVia**, which fundamentally redesigns power delivery. Traditionally, power and data lines shared the front side of the silicon wafer, causing congestion and interference. PowerVia moves the power supply to the backside of the wafer, leaving the front primarily for signals. This separation is intended to stabilize power delivery, particularly vital for high-demand workloads like AI, thus reducing latency, increasing frequency stability, and managing thermal load.
