• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
Science
Report

3D neuromorphic prototype reportedly tunes memory timescales across about three orders of magnitude

An October 5 science roundup says researchers demonstrated motion and ultrasound classification with the lab-scale device.

RY
1 Source, 10h ago, first seen 10h ago

TLDR

An October 5 science roundup says Seoul National University and Yonsei researchers built a 3D neuromorphic prototype with electrically tunable memory timescales spanning about three orders of magnitude. It describes stacked memristors and thin-film transistors used to demonstrate motion and ultrasound classification. Peripheral power use, device variation, endurance and manufacturing yield still need verification; it is not a commercial AI chip.

Combined views

210

1 Source, first seen 10h ago

3 likes1 saves

Combined views

210

1 Source, first seen 10h ago

3 likes1 saves

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Featured Source

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

1 Source

@Goeun_6121Science & Tech Daily | October 5, 2026 Seven developments from October 3 and 4 worth tracking, all from peer-reviewed early releases. 1. Seoul National University and Yonsei researchers built a 3D neuromorphic device with electrically tunable memory timescales spanning about three orders of magnitude. HfO2 memristors stacked with In2O3 thin-film transistors in one vertical process. Circuit configuration stretched the usable range past 10^8; motion and ultrasound classification demonstrated. Reservoir computing trains cheaply but locks to a device's own relaxation time; this design breaks that lock. Lab-scale prototype; peripheral power, device variation, endurance and wafer yield need verification against digital accelerators. Not a commercial AI chip. 2. A multivalent antiviral candidate cross-links influenza virions, blocking entry and release at once. Three neuraminidase inhibitors joined to an antibody Fc fragment to bind multiple viral particles together. The peramivir-based AVC-P3 outperformed baloxavir and CD388 in mouse models of H1N1, H3N2, H5N1 and influenza B. Beyond keeping a drug in place longer: changing valency turns the drug into a physical tether. Preclinical mouse and test-tube work. Human safety, dosing, immune effects, resistance and activity across clinical isolates are untested. Not an approved therapy. 3. A hydrogen plasma reactor turned quartz and rice husks into silicon in two minutes. Passing the feedstock through a 10% hydrogen-argon flight plasma produced crystalline silicon, some particles locally reaching 2N purity. Silicon carbide, partially reduced silica and impurity oxides remained. Could cut carbon-based reduction and multi-stage high-temperature refining from silicon production. 99% is far below solar or semiconductor grade: a proof of concept, not a production process. Yield, power use, impurity removal, economics and lifecycle emissions need testing. 4. A small randomized trial found short-term symptom improvement from tDCS in treatment-resistant schizophrenia. 45 inpatients, frontal-temporal stimulation or sham, twice daily for five days; 39 completed. Active group saw larger PANSS reductions. This adds controlled evidence for non-invasive brain stimulation as an adjunct in patients with limited drug response. Five-day, single-center, 39 completers, and past results have been inconsistent. Durability, functional recovery, drug interactions and multi-site replication remain open. 5. Researchers recovered DNA and RNA information from paraffin-preserved brains stored since the 1940s. Extraction and library methods were compared on formalin-fixed samples from Danish psychiatric institutions. RNA end-repair raised transcript diversity, and protein-coding and microRNA signals came back even from heavily fragmented samples. Archival pathology collections long considered dead for molecular analysis could reopen historical cohorts for neuropsychiatric research. Small-sample feasibility study, not established biomarkers. Reproducibility, contamination, batch effects, record quality and ethical data use need attention. 6. A synthetic bacterial consortium divided labor by chemical signaling to degrade multiple herbicides together. Two engineered consortia split metabolic roles via different quorum-sensing signals, raising co-degradation in lab and soil. Maize pot trials and metagenomics showed native-microbe interaction and soil benefits. Instead of loading every pathway into one strain, microbes communicate and specialize: a bioremediation strategy for mixed herbicide residues. Not a long-term field trial. Stability across soils and climates, ecosystem disturbance, strain persistence and real yields and costs need verification. 7. A small membrane-protein family was tied to C. difficile spore formation. Seven c-di-GMP riboswitch-regulated membrane-protein genes identified. Overexpressing CD1980.2 raised sporulation; deleting all seven sharply cut it. C. difficile spores persist in hospitals and drive transmission and recurrence; mapping how an intracellular signal gates sporulation points to new infection-control targets. Did not treat patients or block transmission. Each protein's direct targets, roles in the human gut, and resistance or side effects from blocking the pathway are unverified. A shared pattern runs through the weekend's work: fixed properties, drug actions and microbial functions are being recast as tunable dynamics. The next test is the same everywhere: large chips, human trials, open-field ecosystems.10h
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    1 Source

    @Goeun_6121Science & Tech Daily | October 5, 2026 Seven developments from October 3 and 4 worth tracking, all from peer-reviewed early releases. 1. Seoul National University and Yonsei researchers built a 3D neuromorphic device with electrically tunable memory timescales spanning about three orders of magnitude. HfO2 memristors stacked with In2O3 thin-film transistors in one vertical process. Circuit configuration stretched the usable range past 10^8; motion and ultrasound classification demonstrated. Reservoir computing trains cheaply but locks to a device's own relaxation time; this design breaks that lock. Lab-scale prototype; peripheral power, device variation, endurance and wafer yield need verification against digital accelerators. Not a commercial AI chip. 2. A multivalent antiviral candidate cross-links influenza virions, blocking entry and release at once. Three neuraminidase inhibitors joined to an antibody Fc fragment to bind multiple viral particles together. The peramivir-based AVC-P3 outperformed baloxavir and CD388 in mouse models of H1N1, H3N2, H5N1 and influenza B. Beyond keeping a drug in place longer: changing valency turns the drug into a physical tether. Preclinical mouse and test-tube work. Human safety, dosing, immune effects, resistance and activity across clinical isolates are untested. Not an approved therapy. 3. A hydrogen plasma reactor turned quartz and rice husks into silicon in two minutes. Passing the feedstock through a 10% hydrogen-argon flight plasma produced crystalline silicon, some particles locally reaching 2N purity. Silicon carbide, partially reduced silica and impurity oxides remained. Could cut carbon-based reduction and multi-stage high-temperature refining from silicon production. 99% is far below solar or semiconductor grade: a proof of concept, not a production process. Yield, power use, impurity removal, economics and lifecycle emissions need testing. 4. A small randomized trial found short-term symptom improvement from tDCS in treatment-resistant schizophrenia. 45 inpatients, frontal-temporal stimulation or sham, twice daily for five days; 39 completed. Active group saw larger PANSS reductions. This adds controlled evidence for non-invasive brain stimulation as an adjunct in patients with limited drug response. Five-day, single-center, 39 completers, and past results have been inconsistent. Durability, functional recovery, drug interactions and multi-site replication remain open. 5. Researchers recovered DNA and RNA information from paraffin-preserved brains stored since the 1940s. Extraction and library methods were compared on formalin-fixed samples from Danish psychiatric institutions. RNA end-repair raised transcript diversity, and protein-coding and microRNA signals came back even from heavily fragmented samples. Archival pathology collections long considered dead for molecular analysis could reopen historical cohorts for neuropsychiatric research. Small-sample feasibility study, not established biomarkers. Reproducibility, contamination, batch effects, record quality and ethical data use need attention. 6. A synthetic bacterial consortium divided labor by chemical signaling to degrade multiple herbicides together. Two engineered consortia split metabolic roles via different quorum-sensing signals, raising co-degradation in lab and soil. Maize pot trials and metagenomics showed native-microbe interaction and soil benefits. Instead of loading every pathway into one strain, microbes communicate and specialize: a bioremediation strategy for mixed herbicide residues. Not a long-term field trial. Stability across soils and climates, ecosystem disturbance, strain persistence and real yields and costs need verification. 7. A small membrane-protein family was tied to C. difficile spore formation. Seven c-di-GMP riboswitch-regulated membrane-protein genes identified. Overexpressing CD1980.2 raised sporulation; deleting all seven sharply cut it. C. difficile spores persist in hospitals and drive transmission and recurrence; mapping how an intracellular signal gates sporulation points to new infection-control targets. Did not treat patients or block transmission. Each protein's direct targets, roles in the human gut, and resistance or side effects from blocking the pathway are unverified. A shared pattern runs through the weekend's work: fixed properties, drug actions and microbial functions are being recast as tunable dynamics. The next test is the same everywhere: large chips, human trials, open-field ecosystems.10h
    Today's Rank

    #13

    Today's Rank

    #13