SetupAI
繁
SourceStalecollected in 61m

Memory chip operates at 700 degrees Celsius

Read original on The Next Web (TNW)
#hardware#edge-ai#semiconductor

New chip tech enables AI compute in extreme heat, opening frontiers for aerospace and industrial AI deployment.

30-Second TL;DR

What Changed

Chip operates at 700°C, far exceeding standard silicon limits

Why It Matters

This breakthrough could revolutionize edge AI in aerospace, deep-earth drilling, and industrial monitoring where heat currently prevents electronics deployment.

What To Do Next

Monitor the startup's progress for potential partnerships if you are building AI applications for industrial or aerospace hardware.

Who should care:Developers & AI Engineers

Key Points

  • •Chip operates at 700°C, far exceeding standard silicon limits
  • •Enables AI computing in extreme environments like Venus
  • •Startup is building specialized AI chips for non-GPU compute

Deep Insight

Background and context from public sources — not the original article. 13 sources cited.

Enhanced Key Takeaways

  • •The startup behind this innovation is TetraMem Inc., with the core memristor technology developed by a research team at the University of Southern California (USC) led by Professor Joshua Yang.
  • •The memristor's extreme temperature resilience is attributed to its unique material composition, featuring a tungsten top electrode, a hafnium oxide ceramic middle layer, and a graphene bottom layer, which prevents atomic migration and short circuits common in high-heat environments.
  • •At 700°C, the device demonstrated impressive performance, including data retention for over 50 hours without needing to be refreshed, endurance exceeding one billion switching cycles, and energy-efficient operation at approximately 1.5 volts.
  • •This memristor design inherently supports in-memory computing, performing critical AI operations like matrix multiplication directly within the memory, which significantly reduces energy consumption and latency by overcoming the traditional von Neumann bottleneck.

Technical Deep Dive

  • Device Type: Memristor (Resistive Random-Access Memory - RRAM), capable of both data storage and computation.
  • Material Stack:
    • Top Electrode: Tungsten, selected for its exceptionally high melting point.
    • Insulating/Switching Layer: Hafnium oxide ceramic.
    • Bottom Electrode: Graphene, a single-atom-thick carbon sheet, which provides mechanical strength, thermal stability, and crucially, prevents metal atom diffusion that typically causes device failure at high temperatures.
  • Operating Principle: Leverages Ohm's Law to perform matrix multiplication physically as electricity flows through the device, enabling in-memory computing for AI workloads.
  • Performance at 700°C:
    • Data Retention: Maintained data for over 50 hours without refresh (with an average of 145 hours across 30 tested devices, ranging from 130 to 170 hours).
    • Endurance: Withstood over one billion switching cycles.
    • Operating Voltage: Approximately 1.5 volts.
    • Switching Speed: Achieved speeds in the tens of nanoseconds (around 30 nanoseconds).
    • Energy Efficiency: At 700°C, the devices required less than one-third of the current and half the voltage for switching compared to room temperature operation, indicating improved energy efficiency at elevated temperatures.
  • Manufacturing Compatibility: Tungsten and hafnium oxide are standard in semiconductor foundries, and graphene is on the development roadmaps of major manufacturers like TSMC and Samsung, indicating potential for industrial scaling despite current manual, small-scale fabrication.

Future ImplicationsAI analysis grounded in cited sources

AI systems will achieve unprecedented autonomy in extreme environments.
The ability to operate memory at 700°C removes a critical thermal barrier, enabling AI computing in previously inaccessible locations such as the surface of Venus, deep-earth drilling sites, and nuclear facilities without the need for complex cooling systems.
Energy consumption for AI computing will be significantly reduced.
The memristor's in-memory computing architecture inherently minimizes data movement, and its improved energy efficiency at high temperatures will reduce the need for power-intensive cooling, leading to substantial energy savings in AI data centers and edge devices.
New frontiers in space exploration and industrial monitoring will open up for advanced intelligent systems.
Robust electronics capable of on-site data processing in harsh conditions will enable more sophisticated scientific missions to extreme celestial bodies and allow for intelligent sensing in challenging industrial settings.

Timeline

2023
TetraMem demonstrated high-density, high-precision RRAM with 2,048 conductance levels in integrated arrays, published in Nature.
2026-03-26
Research on the 700°C memristor by the USC team led by Joshua Yang was published in the journal Science.
2026-04-05
The University of Southern California (USC) announced the graphene memristor breakthrough capable of operating at 700°C.
2026-05-06
Physics World reported on the 700°C memory device, identifying TetraMem Inc. as the startup founded by the USC researchers to commercialize memristor-based AI accelerators.
2026-05-14
TetraMem Inc. officially announced that its academic and research collaborators demonstrated RRAM (memristor) devices capable of reliable operation at temperatures up to 700°C.
Read original article on The Next Web (TNW)

Weekly AI Recap

Read this week's curated digest of top AI events →

PostLinkedIn

AI-curated news aggregator. All content rights belong to original publishers.
Original source: The Next Web (TNW) ↗

This is a summary, not the original. Read the source, or get the weekly briefing.

The weekly digest

One email a week. Unsubscribe anytime.