https://siliconandsystems.com/en/articles/industry-memory-storage-2026/ Silicon & Systems en 2026-08-30 The Boundary Learned to Move: 52 Industry Memory and Storage Papers in 2026 https://siliconandsystems.com/en/articles/vera-cpu-memory-balance/ Silicon & Systems en 2026-08-29 Vera Rebalances the CPU: 88 Cores Fed by 1.2 TB/s https://siliconandsystems.com/en/articles/tencent-pegasus-fabric/ Silicon & Systems en 2026-08-30 One DPU, Eight RNICs: Tencent's Pegasus Network for Bare-Metal AI Cloud https://siliconandsystems.com/en/articles/cxl-onechip/ Silicon & Systems en 2026-08-30 Beyond NVLink and InfiniBand: A One-Chip-Like CXL Datacenter https://siliconandsystems.com/en/articles/burst-aware-congestion-evaluation/ Silicon & Systems en 2026-08-30 A Flow-Size Distribution Cannot Describe a Burst https://siliconandsystems.com/en/articles/industry-circuits-2026/ Silicon & Systems en 2026-08-30 Arrays, Timing, and Structured Compute: Industry Circuits in 2026 https://siliconandsystems.com/en/articles/ai-infrastructure-runtime-2025-2026/ Silicon & Systems en 2026-08-28 The Datacenter Became the Runtime: Nine Industry Papers That Redefined AI Infrastructure https://siliconandsystems.com/en/articles/murakkab-agent-orchestration/ Silicon & Systems en 2026-08-29 The Agent Graph Becomes the Cloud Scheduler: Murakkab Optimizes the Whole Workflow https://siliconandsystems.com/en/articles/strata-kv-cache/ Silicon & Systems en 2026-08-29 A Cache Hit That Arrives Late Is Still a Miss: Strata Schedules Long-Context Memory https://siliconandsystems.com/en/articles/weave-rl-coscheduling/ Silicon & Systems en 2026-08-29 Fill the Idle Half of On-Policy RL: Weave Co-Schedules Rollout and Training https://siliconandsystems.com/en/articles/ampereone-memory-tagging/ Silicon & Systems en 2026-08-30 AmpereOne Hides Memory Tags in the Bits a Server Already Pays For https://siliconandsystems.com/en/articles/m100-dataflow/ Silicon & Systems en 2026-08-30 Li Auto Lets the Compiler Drive Data Instead of Rebuilding a GPU Cache https://siliconandsystems.com/en/articles/sigries-rowhammer/ Silicon & Systems en 2026-08-30 Microsoft's Rowhammer Defense Sleeps Until One Sub-Bank Looks Dangerous https://siliconandsystems.com/en/articles/samsung-3d-stacked-fet/ Silicon & Systems en 2026-08-30 Three Nanosheets over Three: Samsung Builds Logic Upward at 42 nm https://siliconandsystems.com/en/articles/reasoning-inference-capacity/ Silicon & Systems en 2026-08-30 Reasoning Models Hit the Memory-Capacity Wall Before the Compute Ceiling https://siliconandsystems.com/en/articles/matryoshka-network-compiler/ Silicon & Systems en 2026-08-30 A 100K-GPU Network Starts as a Compiler Problem https://siliconandsystems.com/en/articles/meta-matryoshka/ Silicon & Systems en 2026-08-30 Compile the Fabric before Deploying It: Meta's Matryoshka Network Design System https://siliconandsystems.com/en/articles/spec-cpu-2026/ Silicon & Systems en 2026-08-30 SPEC CPU 2026 Stops Pretending Every Core Runs the Same Program https://siliconandsystems.com/en/articles/gpu-cloud-lottery/ Silicon & Systems en 2026-08-30 The Same H100 Is Not the Same Rental: Measuring the GPU Cloud Lottery https://siliconandsystems.com/en/articles/acos-object-store/ Silicon & Systems en 2026-08-30 Five Regions Changed the Economics of Apple's Object Store https://siliconandsystems.com/en/articles/aiturbo-grouped-io/ Silicon & Systems en 2026-08-29 When Storage Understands the Job: AITURBO Turns AI I/O into a Group Operation https://siliconandsystems.com/en/articles/cacheslide-kv-cache/ Silicon & Systems en 2026-08-30 KV Reuse Breaks When the Text Moves, Even If the Text Does Not https://siliconandsystems.com/en/articles/cloud-local-storage/ Silicon & Systems en 2026-08-30 The SSD Stayed Local, but Its Control Path Moved Three Times https://siliconandsystems.com/en/articles/programmable-page-cache-model-loading/ Silicon & Systems en 2026-08-30 The Fastest Model Loader Changed the Page Cache, Not the Framework https://siliconandsystems.com/en/articles/sorted-compressed-filesystem/ Silicon & Systems en 2026-08-30 Compression Improves When the Filesystem Sorts Before It Packs https://siliconandsystems.com/en/articles/tape-archive-cloud/ Silicon & Systems en 2026-08-30 A Thousand Cartridges, Four Drives: Why Tape Cloud Storage Must Wait on Purpose https://siliconandsystems.com/en/articles/kernelevolve/ Silicon & Systems en 2026-08-30 Meta Gives Kernel Optimization a Search Tree and a Memory https://siliconandsystems.com/en/articles/tsmc-cfet-circuits/ Silicon & Systems en 2026-08-30 CFET Leaves the Device Lab: TSMC Runs an Oscillator and SRAM below 48 nm https://siliconandsystems.com/en/articles/industry-memory-storage-2025/ Silicon & Systems en 2026-08-30 The Hierarchy Became the Product: 34 Industry Memory and Storage Papers from 2025 https://siliconandsystems.com/en/articles/meta-100k-gpu-network/ Silicon & Systems en 2026-08-30 One RoCE Fabric, Four Distance Classes: Meta's Communication Stack for 100K+ GPUs https://siliconandsystems.com/en/articles/ecoc-space-wavelength-switch/ Silicon & Systems en 2026-08-30 One Photonic Switch, Two Routing Dimensions https://siliconandsystems.com/en/articles/mitsubishi-400g-eml/ Silicon & Systems en 2026-08-30 At 100 GHz, the Laser Package Becomes Part of the Circuit https://siliconandsystems.com/en/articles/bytetracker-agentless-probing/ Silicon & Systems en 2026-08-30 Ask the Switch Which Path the Probe Actually Took https://siliconandsystems.com/en/articles/industry-circuits-2025/ Silicon & Systems en 2026-08-30 Precision Moves Into the Architecture: Industry Circuits in 2025 https://siliconandsystems.com/en/articles/sglb-global-load-balancing/ Silicon & Systems en 2026-08-30 A Local Queue Cannot See the Congestion Two Switches Away https://siliconandsystems.com/en/articles/skeletonhunter/ Silicon & Systems en 2026-08-30 The Training Job Already Knows Which Network Paths Matter https://siliconandsystems.com/en/articles/ayar-ucie-optical-retimer/ Silicon & Systems en 2026-08-30 UCIe Crosses the Rack in Light: Ayar Labs' 8.192 Tb/s Optical Retimer https://siliconandsystems.com/en/articles/celestial-photonic-fabric/ Silicon & Systems en 2026-08-30 Optics Moves Inside the Silicon: Celestial AI's Photonic Fabric Module https://siliconandsystems.com/en/articles/lightmatter-passage-m1000/ Silicon & Systems en 2026-08-30 A 4,000 mm² Optical Backplane: Reading Lightmatter's Passage M1000 https://siliconandsystems.com/en/articles/greyhound-fail-slow/ Silicon & Systems en 2026-08-30 The GPU Did Not Crash, Yet 1,024 Workers Slowed Down https://siliconandsystems.com/en/articles/torpor-serverless-gpu/ Silicon & Systems en 2026-08-30 When Idle Models Leave the GPU: Paying Only for Active Inference https://siliconandsystems.com/en/articles/dcperf-cpu-benchmark/ Silicon & Systems en 2026-08-30 A CPU Benchmark That Meta Was Willing to Buy Servers With https://siliconandsystems.com/en/articles/llama3-parallelism/ Silicon & Systems en 2026-08-30 Why Llama 3 Needed Four Kinds of Parallelism at Once https://siliconandsystems.com/en/articles/c4-training-efficiency/ Silicon & Systems en 2026-08-30 Alibaba Makes Collective Communication Diagnose the Cluster https://siliconandsystems.com/en/articles/meta-ml-reliability/ Silicon & Systems en 2026-08-30 The Largest GPU Jobs Fail Most, but Small Jobs Still Set Fleet Policy https://siliconandsystems.com/en/articles/mlperf-power/ Silicon & Systems en 2026-08-30 AI Energy Needs a Meter That Survives Nine Orders of Magnitude https://siliconandsystems.com/en/articles/ibm-telum-ii/ Silicon & Systems en 2026-08-30 A 600 mm² Mainframe Die Becomes a System: Inside IBM Telum II https://siliconandsystems.com/en/articles/lumorph-photonic-rack/ Silicon & Systems en 2026-08-30 Let the Optical Topology Follow the GPU Allocation https://siliconandsystems.com/en/articles/industry-circuits-2024/ Silicon & Systems en 2026-08-30 What Industry Put on Silicon in 2024 https://siliconandsystems.com/en/articles/industry-memory-storage-2024/ Silicon & Systems en 2026-08-30 Memory Stopped Being a Component: 35 Industry Papers from 2024 https://siliconandsystems.com/en/articles/mi300a-package/ Silicon & Systems en 2026-08-30 MI300A Turns the Accelerator Package into the Compute Node https://siliconandsystems.com/en/articles/tensor-contraction-processor/ Silicon & Systems en 2026-08-30 FuriosaAI Replaces the Fixed Matrix with a Shape-Shifting Tensor Engine https://siliconandsystems.com/en/articles/traffic-reprofiling-deadlines/ Silicon & Systems en 2026-08-30 Spend Delay Before the First Hop to Buy Back Bandwidth https://siliconandsystems.com/ko/articles/industry-memory-storage-2026/ Silicon & Systems ko 2026-08-30 데이터 경계가 움직이기 시작했습니다: 2026년 산업계 메모리·스토리지 논문 52편 https://siliconandsystems.com/ko/articles/vera-cpu-memory-balance/ Silicon & Systems ko 2026-08-29 Vera CPU의 설계 기준: 88개 코어에 1.2TB/s를 배분한 이유 https://siliconandsystems.com/ko/articles/tencent-pegasus-fabric/ Silicon & Systems ko 2026-08-30 DPU 하나와 RNIC 여덟 개, Tencent의 베어메탈 AI 클라우드 네트워크 https://siliconandsystems.com/ko/articles/cxl-onechip/ Silicon & Systems ko 2026-08-30 NVLink와 InfiniBand의 경계를 넘어: 원칩형 CXL 데이터센터 https://siliconandsystems.com/ko/articles/burst-aware-congestion-evaluation/ Silicon & Systems ko 2026-08-30 플로 크기 분포만으로는 버스트를 설명할 수 없습니다 https://siliconandsystems.com/ko/articles/industry-circuits-2026/ Silicon & Systems ko 2026-08-30 배열과 타이밍, 데이터 구조를 함께 설계한 2026년 회로 https://siliconandsystems.com/ko/articles/ai-infrastructure-runtime-2025-2026/ Silicon & Systems ko 2026-08-28 데이터센터 전체가 런타임이 되었습니다: 산업계 논문 9편으로 읽는 AI 인프라 https://siliconandsystems.com/ko/articles/murakkab-agent-orchestration/ Silicon & Systems ko 2026-08-29 에이전트 워크플로 전체를 스케줄링하는 클라우드: Murakkab https://siliconandsystems.com/ko/articles/strata-kv-cache/ Silicon & Systems ko 2026-08-29 장문맥 추론의 병목은 캐시 용량보다 도착 시간입니다 https://siliconandsystems.com/ko/articles/weave-rl-coscheduling/ Silicon & Systems ko 2026-08-29 롤아웃과 학습 사이의 빈 시간을 회수하는 Weave https://siliconandsystems.com/ko/articles/ampereone-memory-tagging/ Silicon & Systems ko 2026-08-30 서버가 이미 가진 보호 비트에 메모리 태그를 넣은 AmpereOne https://siliconandsystems.com/ko/articles/m100-dataflow/ Silicon & Systems ko 2026-08-30 GPU식 캐시 대신 컴파일러로 데이터를 움직이는 자동차 AI 칩 https://siliconandsystems.com/ko/articles/sigries-rowhammer/ Silicon & Systems ko 2026-08-30 평상시에는 접근 횟수를 세고, 위험한 메모리 구역만 샘플링하는 Rowhammer 방어 https://siliconandsystems.com/ko/articles/samsung-3d-stacked-fet/ Silicon & Systems ko 2026-08-30 42 nm에서 위로 쌓은 로직, 삼성의 3단 나노시트 적층 FET https://siliconandsystems.com/ko/articles/reasoning-inference-capacity/ Silicon & Systems ko 2026-08-30 추론 연산량보다 먼저 메모리 용량 한계에 닿는 추론형 모델 https://siliconandsystems.com/ko/articles/matryoshka-network-compiler/ Silicon & Systems ko 2026-08-30 10만 GPU 네트워크는 컴파일러에서 시작됩니다 https://siliconandsystems.com/ko/articles/meta-matryoshka/ Silicon & Systems ko 2026-08-30 배포 전에 패브릭을 컴파일하는 Meta의 Matryoshka https://siliconandsystems.com/ko/articles/spec-cpu-2026/ Silicon & Systems ko 2026-08-30 SPEC CPU 2026은 모든 코어가 같은 프로그램을 실행한다는 가정을 버렸습니다 https://siliconandsystems.com/ko/articles/gpu-cloud-lottery/ Silicon & Systems ko 2026-08-30 같은 H100을 빌려도 같은 성능이 나오지 않습니다 https://siliconandsystems.com/ko/articles/acos-object-store/ Silicon & Systems ko 2026-08-30 리전 다섯 곳이 Apple 오브젝트 스토리지의 비용 구조를 바꿨습니다 https://siliconandsystems.com/ko/articles/aiturbo-grouped-io/ Silicon & Systems ko 2026-08-29 AI 스토리지가 작업 의도를 읽는 법: AITURBO의 그룹 I/O https://siliconandsystems.com/ko/articles/cacheslide-kv-cache/ Silicon & Systems ko 2026-08-30 문장이 같아도 위치가 바뀌면 KV 캐시는 그대로 쓸 수 없습니다 https://siliconandsystems.com/ko/articles/cloud-local-storage/ Silicon & Systems ko 2026-08-30 SSD는 서버에 남았지만 제어 경로는 세 번 바뀌었습니다 https://siliconandsystems.com/ko/articles/programmable-page-cache-model-loading/ Silicon & Systems ko 2026-08-30 모델 로딩을 줄인 핵심은 프레임워크가 아니라 페이지 캐시 정책입니다 https://siliconandsystems.com/ko/articles/sorted-compressed-filesystem/ Silicon & Systems ko 2026-08-30 압축 전에 데이터를 정렬하면 파일시스템의 밀도와 읽기가 함께 개선됩니다 https://siliconandsystems.com/ko/articles/tape-archive-cloud/ Silicon & Systems ko 2026-08-30 카트리지 1,000개와 드라이브 4개: 테이프 클라우드가 기다리는 이유 https://siliconandsystems.com/ko/articles/kernelevolve/ Silicon & Systems ko 2026-08-30 커널 최적화에 탐색 트리와 기억을 붙인 Meta의 에이전트 https://siliconandsystems.com/ko/articles/tsmc-cfet-circuits/ Silicon & Systems ko 2026-08-30 48 nm 아래에서 회로가 된 CFET, TSMC의 링 오실레이터와 SRAM https://siliconandsystems.com/ko/articles/industry-memory-storage-2025/ Silicon & Systems ko 2026-08-30 계층 전체가 제품이 되었습니다: 2025년 산업계 메모리·스토리지 논문 34편 https://siliconandsystems.com/ko/articles/meta-100k-gpu-network/ Silicon & Systems ko 2026-08-30 하나의 RoCE 패브릭에 네 가지 거리, Meta의 10만 GPU 통신 스택 https://siliconandsystems.com/ko/articles/ecoc-space-wavelength-switch/ Silicon & Systems ko 2026-08-30 공간과 파장을 함께 고르는 4×4 광 스위치 https://siliconandsystems.com/ko/articles/mitsubishi-400g-eml/ Silicon & Systems ko 2026-08-30 100GHz를 넘으면 레이저 패키지도 RF 회로가 됩니다 https://siliconandsystems.com/ko/articles/bytetracker-agentless-probing/ Silicon & Systems ko 2026-08-30 장애 패킷이 실제로 지난 경로를 기록하는 ByteTracker https://siliconandsystems.com/ko/articles/industry-circuits-2025/ Silicon & Systems ko 2026-08-30 정밀도를 구조 안으로 옮긴 2025년 산업계 회로 https://siliconandsystems.com/ko/articles/sglb-global-load-balancing/ Silicon & Systems ko 2026-08-30 로컬 큐만으로는 원격 경로의 혼잡을 알 수 없습니다 https://siliconandsystems.com/ko/articles/skeletonhunter/ Silicon & Systems ko 2026-08-30 학습 작업이 중요한 네트워크 경로를 이미 알려 줍니다 https://siliconandsystems.com/ko/articles/ayar-ucie-optical-retimer/ Silicon & Systems ko 2026-08-30 빛으로 랙을 건너는 UCIe, Ayar Labs의 8.192 Tb/s 광 리타이머 https://siliconandsystems.com/ko/articles/celestial-photonic-fabric/ Silicon & Systems ko 2026-08-30 실리콘 안으로 들어온 광 I/O, Celestial AI의 포토닉 패브릭 모듈 https://siliconandsystems.com/ko/articles/lightmatter-passage-m1000/ Silicon & Systems ko 2026-08-30 4,000 mm² 광 백플레인, Lightmatter Passage M1000의 구조와 한계 https://siliconandsystems.com/ko/articles/greyhound-fail-slow/ Silicon & Systems ko 2026-08-30 멈추지 않은 GPU 하나가 대규모 학습을 늦출 때 https://siliconandsystems.com/ko/articles/torpor-serverless-gpu/ Silicon & Systems ko 2026-08-30 요청이 올 때만 GPU를 쓰는 서버리스 추론 https://siliconandsystems.com/ko/articles/dcperf-cpu-benchmark/ Silicon & Systems ko 2026-08-30 실제 서버 구매에 쓰인 CPU 벤치마크 https://siliconandsystems.com/ko/articles/llama3-parallelism/ Silicon & Systems ko 2026-08-30 Llama 3 학습에 네 가지 병렬화가 동시에 필요했던 이유 https://siliconandsystems.com/ko/articles/c4-training-efficiency/ Silicon & Systems ko 2026-08-30 집합 통신으로 학습 클러스터의 장애까지 진단하다 https://siliconandsystems.com/ko/articles/meta-ml-reliability/ Silicon & Systems ko 2026-08-30 GPU 학습 장애를 대형 작업만으로 설명할 수 없는 이유 https://siliconandsystems.com/ko/articles/mlperf-power/ Silicon & Systems ko 2026-08-30 마이크로와트에서 메가와트까지 통하는 AI 전력 측정법 https://siliconandsystems.com/ko/articles/ibm-telum-ii/ Silicon & Systems ko 2026-08-30 600 mm² 메인프레임 칩 안에 시스템을 넣은 IBM Telum II https://siliconandsystems.com/ko/articles/lumorph-photonic-rack/ Silicon & Systems ko 2026-08-30 GPU 배치에 맞춰 광 토폴로지를 바꾸는 랙 https://siliconandsystems.com/ko/articles/industry-circuits-2024/ Silicon & Systems ko 2026-08-30 2024년 산업계 회로 논문이 실리콘으로 증명한 것 https://siliconandsystems.com/ko/articles/industry-memory-storage-2024/ Silicon & Systems ko 2026-08-30 부품이 아니라 배치 문제였습니다: 2024년 산업계 메모리·스토리지 논문 35편 https://siliconandsystems.com/ko/articles/mi300a-package/ Silicon & Systems ko 2026-08-30 MI300A는 가속기 패키지를 하나의 컴퓨팅 노드로 바꿨습니다 https://siliconandsystems.com/ko/articles/tensor-contraction-processor/ Silicon & Systems ko 2026-08-30 고정 행렬을 버리고 텐서 모양에 맞춰 나뉘는 가속기 https://siliconandsystems.com/ko/articles/traffic-reprofiling-deadlines/ Silicon & Systems ko 2026-08-30 첫 홉 전에 지연을 쓰면 필요한 대역폭을 줄일 수 있습니다