NHanced Semiconductors and University of Florida to Present Advanced Photonics Packaging Research

thequantuminsider.com Aug 18, 2026

NHanced Semiconductors and University of Florida to Present Advanced Photonics Packaging Research

Curated by Shubham · TechCapsules · Aug 18, 2026

AI-summarised brief · reviewed before publication

NHanced Semiconductors and the University of Florida will present a paper on advanced packaging for silicon photonics at the 2026 SMTA International Conference. The study focuses on copper‑copper hybrid bonding and silicon interposers for 2.5D heterogeneous integration of chiplets. It addresses bonding defects, alignment accuracy, thermal expansion, and heat dissipation in densely packed photonic‑electronic systems. The presentation, led by Dr. Charles Woychik and Dr. Navid Asadizanjani, highlights fine‑pitch Cu‑Cu bonding and high‑density silicon interposers as key enablers for high‑bandwidth, energy‑efficient photonic systems.

💡 Why It Matters

· The research offers a practical roadmap for overcoming packaging bottlenecks that have stalled the scaling of silicon photonics, directly impacting the performance of next‑generation data‑center interconnects.

Capsules Semiconductors

Read full article on thequantuminsider.com

Cerebras CS-4 rack systems juice their dinner-plate-sized AI chips for every last…

Cerebras unveiled its next-generation Wafer Scale Engine (WSE-3T) and Nexus rack systems, designed to maximize AI inference performance through enhanced power delivery rather than new silicon. The WSE-3T utilizes the same TSMC 5nm process, 4 trillion transistors, and 46,225 mm² wafer area as its predecessor but doubles compute, memory fabric, and I/O bandwidth. By increasing the thermal design power from 15 kW to an estimated 33 kW per wafer, the chip operates at approximately 2.8 GHz, up from 1.4 GHz. This results in 250 PFLOPS of sparse FP16 compute and 43.2 PB/s of memory bandwidth. Cerebras claims this approach extends its lead over competitors like Nvidia and AMD. The new architecture allows racks to house three times as many chips while doubling per-chip performance. The company emphasizes that efficient power delivery is the primary innovation enabling these higher operating frequencies and faster token generation rates for high-speed AI workloads.

💡 Cerebras bypasses the industry's reliance on shrinking process nodes by extracting double the performance from existing silicon through aggressive power scaling. This strategy challenges the conventional wisdom that hardware upgrades require new manufacturing technology, potentially reshaping how AI infrastructure efficiency is measured.

nynewscast.com · Aug 19

Cerebras Unveils CS-4: Up to 30 Times Faster than GPU-based Solutions

Cerebras Systems introduced the Cerebras CS-4, a rack-scale AI accelerator claiming to be the industry’s fastest. Built on the new Nexus platform architecture, the CS-4 utilizes three Wafer Scale Engine 3 Turbo units. It delivers 750 PFLOPs of AI compute and 129.6 petabytes per second of memory bandwidth. The system achieves up to twice the speed of the previous CS-3 model. Consequently, it offers up to 30 times faster token generation per user compared to GPU-based solutions. Additionally, the CS-4 provides up to 10 times more throughput per watt than its predecessor. This efficiency aims to improve data center economics by delivering higher-value tokens within existing power budgets. The hardware supports models exceeding 50 trillion parameters with ultra-low latency. Industry experts note significant improvements in deployability and networking. The CS-4 is designed to scale performance for large-scale token factories and frontier models. It represents a major shift in AI inference speed and capacity for production workloads.

💡 The hardware enables agentic systems to perform an order of magnitude more reasoning and verification within the same timeframe. This speed transforms AI from a reactive tool into a highly capable, real-time production engine.

theregister.com · Aug 19

Cerebras CS-4 rack systems juice chips for every last drop of AI…

Cerebras unveiled its next-generation Wafer Scale Engine-3T (WSE-3T) and Nexus rack systems, designed to maximize AI inference performance through enhanced power delivery rather than new silicon. The WSE-3T utilizes the existing TSMC 5nm process and 4 trillion transistors but doubles compute, memory fabric, and I/O bandwidth by increasing operating frequencies to an estimated 2.8 GHz. This "Turbo" variant delivers 250 petaFLOPS of sparse FP16 compute and 43.2 PB/s of memory bandwidth, doubling the throughput of the previous WSE-3. However, the system’s thermal design power rises to an estimated 46 kW. While the headline figures appear superior to competitors, the performance relies heavily on sparsity, which offers limited benefits for large language model inference. Consequently, the dense FP16 performance is estimated at 25 petaFLOPS. Cerebras claims this approach extends its lead in memory bandwidth, aiming for tenfold throughput per watt improvements over the prior generation.

💡 Cerebras proves that architectural efficiency can outpace raw transistor scaling, challenging the industry's reliance on newer process nodes for performance gains. By maximizing existing silicon through advanced power delivery, the company demonstrates a viable path to high-throughput inference without the immediate costs of next-generation manufacturing.

theregister.com · Aug 19

Cerebras CS-4 rack systems juice chips for every last drop of AI…

Cerebras unveiled its next-generation Wafer Scale Engine (WSE-3T) and Nexus rack systems, aiming to double per-chip performance while tripling chip density within a single rack. The WSE-3T utilizes existing TSMC 5nm silicon but achieves twice the compute, memory fabric, and I/O bandwidth of the previous WSE-3 model through enhanced power delivery. This innovation allows the chip to operate at an estimated 2.8 GHz, up from 1.4 GHz, delivering 250 petaFLOPS of sparse FP16 compute and 43.2 PB/s of memory bandwidth. While the headline figures appear superior to Nvidia and AMD GPUs, Cerebras’ performance relies heavily on sparsity, which offers limited benefits for large language model inference. Consequently, the dense FP16 performance is estimated at 25 petaFLOPS. The system’s total power draw increases to an estimated 46 kW. Cerebras claims these improvements enable faster token generation and significantly higher throughput per watt, extending its lead in high-speed AI inference despite theoretical limitations in real-world saturation.

💡 Cerebras proves that architectural efficiency and power delivery can outpace raw transistor scaling. By maximizing existing silicon, they challenge the industry's reliance on newer process nodes for performance gains.

siliconangle.com · Aug 19

Inference chip startup Etched raises another $700M at $21B valuation

Inference chip startup Etched Inc. has secured an additional $700 million in funding, led by Jane Street, with participation from Kleiner Perkins, Sequoia, Andreessen Horowitz, Tiger Global, and Bain Capital Ventures. This latest round more than doubles the company’s valuation to $21 billion, following a recent $300 million investment backed by Nvidia Corp. Founded in 2021, Etched pivoted to creating inference accelerators supporting various model architectures. Its first prototype, manufactured by Taiwan Semiconductor Manufacturing Co., recently entered production. Jane Street became the first customer, installing an Etched-powered rack in its data center last month. Etched also operates a 2-megawatt AI cluster in San Jose for prospective clients. The startup claims its chips operate at under half the voltage of competitors, enabling higher FLOPs density. Additionally, Etched’s custom racks feature an internal interconnect that reportedly completes certain communication tasks in 700 milliseconds, compared to 4,000 milliseconds for rival chips.

💡 Etched’s rapid valuation surge and early adoption by a major financial firm validate the market’s shift toward specialized, energy-efficient hardware for AI inference. By demonstrating significant performance gains through low-voltage architecture and faster interconnects, the startup challenges established players to rethink efficiency standards in data center deployments.

techcrunch.com · Aug 19

Etched’s valuation doubles to $21B in a month

Etched announced a $700 million funding round that lifted its valuation to $21 billion, a jump from $10.3 billion in July and $5 billion in December. Jane Street led the investment, following a successful test of Etched’s AI inference hardware. The company’s new prefill chip and cluster‑scale memory aim to accelerate both the prefill and decode stages of AI inference, promising higher speeds and lower costs. Etched’s systems can run any frontier model, dispelling earlier concerns of model‑specific design.

💡 The rapid valuation surge signals a shift in AI hardware, positioning Etched as a viable alternative to Nvidia’s dominant AI factories. Its technology could redefine inference efficiency, influencing how data centers deploy and scale large language models.