Cracking under pressure: silicon fractures on ATLAS strip tracker petals

10 hours ago

The ATLAS experiment will upgrade its Inner Detector to an all-silicon Inner Tracker (ITk) for the High-Luminosity LHC, consisting of pixel and strip detectors with barrel and end-cap regions using staves and petals.
Silicon sensors in modules can crack at temperatures below -35°C, particularly during cooling system failures, rendering them inoperable, which was a critical issue for the ITk project.
Mitigation strategies for petals focus on modifying adhesive choice and deposition patterns for module assembly and petal loading.
The best strategy prevents cracks down to -45°C, with only a small percentage observed at -55°C after cycling, addressing expected cooling system problems and catastrophic failures.

The Intel 386's I/O pins face three major dangers: static electricity, latchup, and metastability, requiring specialized protection circuits.
Protection against static electricity uses diodes and resistors to divert excess voltage and limit current, preventing damage to the thin oxide layer.
Latchup is caused by parasitic transistors forming a feedback loop; guard rings around I/O transistors block unwanted currents and prevent short circuits.

Hidden system prompts can shift an LLM's behavioral fingerprint up to a Jensen-Shannon distance of 0.46, effectively making it behave like a different model.
The study tested 5,660 controlled API calls, injecting real CLI prompts (Codex CLI and Claude Code) into models like GPT-5.5 and gpt-4o-mini, and measured drift in answer distributions.
The Claude Code prompt pushed both models into 'different model' territory: colors collapsed to a single answer, and languages switched (e.g., 30/40 responses in Chinese).

The author prefers real data for circuit illustrations in their book, avoiding fake or retraced diagrams.
Measuring V-I plots for semiconductors is challenging due to tiny currents, sudden spikes causing device failure, and temperature-related drift.
Using a benchtop multimeter and pulsed power helps measure microamps/microvolts and reduce heating effects; cooling with non-conductive liquid like mineral oil is recommended.

The transition from frontier model APIs to building your own open-source LLM inference stack often starts due to financial, legal, or engineering concerns.
The initial decision point is whether to buy from managed inference providers or build with open weights and your own hardware, with this post focusing on the build path.
Starting with a single replica using serving engines like vLLM, SGLang, or TensorRT-LLM typically works well, handling batching and KV cache management effectively.

Distillation of frontier AI models is cheap and hard to stop, threatening labs' business models.
Hiding chain of thought may not prevent distillation, especially for agentic tool use.
Companies can distill models by using user-approved sequences (e.g., gold diffs) as RL training targets.

The Intel 386's I/O pins face three major dangers: static electricity, latchup, and metastability, requiring specialized protection circuits.
Protection against static electricity uses diodes and resistors to divert excess voltage and limit current, preventing damage to the thin oxide layer.
Latchup is caused by parasitic transistors forming a feedback loop; guard rings around I/O transistors block unwanted currents and prevent short circuits.

Hidden system prompts can shift an LLM's behavioral fingerprint up to a Jensen-Shannon distance of 0.46, effectively making it behave like a different model.
The study tested 5,660 controlled API calls, injecting real CLI prompts (Codex CLI and Claude Code) into models like GPT-5.5 and gpt-4o-mini, and measured drift in answer distributions.
The Claude Code prompt pushed both models into 'different model' territory: colors collapsed to a single answer, and languages switched (e.g., 30/40 responses in Chinese).

The author prefers real data for circuit illustrations in their book, avoiding fake or retraced diagrams.
Measuring V-I plots for semiconductors is challenging due to tiny currents, sudden spikes causing device failure, and temperature-related drift.
Using a benchtop multimeter and pulsed power helps measure microamps/microvolts and reduce heating effects; cooling with non-conductive liquid like mineral oil is recommended.

The transition from frontier model APIs to building your own open-source LLM inference stack often starts due to financial, legal, or engineering concerns.
The initial decision point is whether to buy from managed inference providers or build with open weights and your own hardware, with this post focusing on the build path.
Starting with a single replica using serving engines like vLLM, SGLang, or TensorRT-LLM typically works well, handling batching and KV cache management effectively.

Distillation of frontier AI models is cheap and hard to stop, threatening labs' business models.
Hiding chain of thought may not prevent distillation, especially for agentic tool use.
Companies can distill models by using user-approved sequences (e.g., gold diffs) as RL training targets.

Hasty Briefsbeta