Cloud Revenue Soars 20% & AI Models Challenge BERT: This Week’s Tech Highlights

Cloud Services Revenue Grows 20% in H1

According to Synergy Research, global cloud migration in the first half of 2020 increased revenue by 20% compared with the previous year. Revenue across four key cloud services and infrastructure segments reached $187 billion, and analysts predict annual cloud spending will double within four years.

Google Introduces pQRNN Model Matching BERT with Fewer Parameters

Google unveiled the pQRNN model, an evolution of the previous PRADO architecture that combines a simple mapping with a quasi‑RNN encoder for fast parallel processing. Despite using far fewer parameters, pQRNN achieves text‑classification performance comparable to BERT.

China Issues Its First NFC‑Enabled Chip Stamp

China Post released a commemorative stamp that embeds an NFC chip readable via the China Post app. The ultra‑thin 120 µm chip uses high‑level encryption to protect stored data, demonstrating a novel application of NFC technology in philately.

TSMC Plans 3nm Monthly Capacity of 100,000 Wafer‑s in 2023

TSMC, a leader in advanced semiconductor processes, aims to increase its 3nm monthly capacity to 100,000 wafers by 2023, following a risk‑run pilot in 2021 and mass production in late 2022. Current plans target 55,000 wafers per month for the 3nm node.

Intel Launches Processors Optimized for IoT, Edge AI, and 5G

Intel announced a portfolio of processors—including 11th‑gen Core, Atom x6000E, Pentium, Celeron N, and J series—designed to enhance edge computing, AI, IoT, 5G, and security capabilities, addressing challenges such as edge complexity, cost, and environmental impact.

Apple Open‑Sources Swift System with Linux Support

Apple engineers released Swift System, a low‑level system‑interface library, as open source and added preliminary Linux support. While Swift System provides Swift‑native APIs for system calls, full parity with all Linux syscalls is still pending.

FGPM: A Fast Gradient Projection Method for Text Adversarial Samples

The paper proposes FGPM, a synonym‑replacement based adversarial attack that generates text perturbations efficiently. Unlike image‑based attacks, FGPM addresses the unique challenges of preserving semantic meaning while increasing model robustness.

ICLR 2020: Visual Information Boosts Machine Translation

The study introduces a universal visual representation that incorporates image data into machine‑translation models without requiring bilingual image annotations, yielding significant performance gains across multiple datasets.