A few of the ideas, topics, and commonplaces that have been gaining steam on arXiv during the last few months (explainer)
1. RTL generation: RTL refers to Register-transfer level, a relatively high-level abstraction in hardware description languages. Much like in software, you can't really do any sort of complex hardware design work without going up in the abstaction ladder and then using automated tools to fill in the details.
We're in 2025, so naturally there's a minor explosion of work trying to use LLMs as tools for RTL generation. It's not something I'm very optimistic about — surely hardware design is one area where training experts in domain-specific languages has positive payoffs — but I'm listing these papers anyway because I do expect post-LLM AI tools to have a large impact in RTL generation/complex hardware design in general, although less thru natural language than thru higher-level formal languages and more sophisticated and effective compilers to lower-level languages. Either way, it's a god idea to keep an eye open in this direction.
Some recent articles:
- Spec2RTL-Agent: Automated Hardware Code Generation from Complex Specifications Using LLM Agent Systems
- Comprehensive Verilog Design Problems: A Next-Generation Benchmark Dataset for Evaluating Large Language Models and Agents on RTL Design and Verification
- TuRTLe: A Unified Evaluation of LLMs for RTL Generation
- RTL++: Graph-enhanced LLM for RTL Code Generation
- VeriReason: Reinforcement Learning with Testbench Feedback for Reasoning-Enhanced Verilog Generation
2. Contextual metadata: A hodgepodge of fields, but it's always worth looking beyond "data" as a fungible blob.
Some recent articles:
- Time-Aware Auto White Balance in Mobile Photography
- Atomizer: Generalizing to new modalities by breaking satellite images down to a set of scalars
- Modeling Beyond MOS: Quality Assessment Models Must Integrate Context, Reasoning, and Multimodality
- URLs Help, Topics Guide: Understanding Metadata Utility in LLM Training
3. MAGE: I'm keeping this term as a self-reminder of the pitfalls of the sort of language processing I do in this project. In the context of the papers below it refers to:
- an spectrograph
- a physiological metric
- a form of deep-learning embedding
- a gravitational wave detection experiment
- an image synthesis method
A "hot" term? No. Interesting papers? Perhaps! Take a look.
Some recent articles:
- The OVz stars in NGC 346: distribution, stellar parameters and insights into low-metallicity effects
- Maximal Speed of Glucose Change Significantly Distinguishes Prediabetes from Diabetes
- FOLIAGE: Towards Physical Intelligence World Models Via Unbounded Surface Evolution
- Experimental Exclusion of Planetary Mass Primordial Black Hole Mergers
- Plug-and-Play Context Feature Reuse for Efficient Masked Generation