{"@context":"https://neupai.io/schema/v0.2","@type":"StructuredNewsArticle","identity":{"article_id":"skhynix_20261001_ai-ecosystem-infra-redesign","canonical_url":"https://news.skhynix.co.kr/ai-ecosystem-series-ep3/","ai_url":null,"publisher":{"name":"SK하이닉스 뉴스룸","domain":"skhynix.co.kr","type":"online"},"author":"unknown","published_at":"2026-10-01T23:59:00.000Z","updated_at":null,"language":"en","article_type":"feature","originality":"aggregated"},"content":{"headline":"[AI Ecosystem] Infrastructure Redesign: Structure Determines Performance","summary":"As AI workloads become more sophisticated, data movement has emerged as the core bottleneck in system performance, driving demand for a paradigm shift from processor-centric architectures to memory-centric computing, near-memory accelerators, and disaggregated architectures.","topics":["AI infrastructure","memory semiconductors","data centers","system architecture","HBM"],"geography":["KR","CH"],"entities":[{"name":"SK Hynix","canonical_id":"corp:kr:sk-hynix","type":"company","role_in_article":"primary_subject","metadata":{"ticker":"000660.KS","parent":null}},{"name":"Onur Mutlu","canonical_id":"person:ch:onur-mutlu","type":"person","role_in_article":"source","metadata":{}},{"name":"ETH Zurich","canonical_id":"org:ch:eth-zurich","type":"organization","role_in_article":"mentioned","metadata":{}},{"name":"NVIDIA","canonical_id":"corp:us:nvidia","type":"company","role_in_article":"mentioned","metadata":{"ticker":"NVDA"}}],"claims":[{"id":"c1","statement":"The energy required to read data once from DRAM can be 150 to 2,000 times greater than that required for a simple arithmetic operation","as_of":"2026","as_of_explicit":false,"as_of_raw":"2026","source_type":"research_paper","comparison":"arithmetic_operation","type":"fact","figures":{"value":150,"unit":"ratio","approximate":false,"converted":{"value":2000,"unit":"ratio"}},"expiry_hint":null,"insight":null},{"id":"c2","statement":"For large machine learning models, more than 90% of system energy was found to be consumed not by computation but by memory access and data movement","as_of":"2026","as_of_explicit":false,"as_of_raw":"2026","source_type":"research_paper","comparison":null,"type":"fact","figures":{"value":90,"unit":"%","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c3","statement":"The Tesseract study showed approximately 10-fold performance improvements and similar levels of energy savings across various parallel graph processing workloads compared to existing processor-centric designs","as_of":"2015","as_of_explicit":true,"as_of_raw":"ISCA 2015","source_type":"research_paper","comparison":"processor_centric_design","type":"fact","figures":{"value":10,"unit":"ratio","approximate":true,"converted":null},"expiry_hint":null,"insight":null},{"id":"c4","statement":"A CXL-based LLM inference study (the CENT design) presented at ASPLOS 2025 significantly reduced memory capacity and bandwidth burdens in large-scale inference infrastructure, and also delivered improvements in performance, energy, cost, and chip area","as_of":"2025","as_of_explicit":true,"as_of_raw":"ASPLOS 2025","source_type":"research_paper","comparison":null,"type":"fact","figures":null,"expiry_hint":null,"insight":null},{"id":"c5","statement":"It was confirmed that existing DRAM chips can be utilized for certain computations beyond simple data storage","as_of":"2024","as_of_explicit":true,"as_of_raw":"HPCA 2024","source_type":"research_paper","comparison":null,"type":"fact","figures":null,"expiry_hint":null,"insight":null}],"ai_emotional_context":{"valence":0.2,"arousal":0.4,"primary_emotions":[{"emotion":"optimistic","intensity":0.5},{"emotion":"calm","intensity":0.6}],"secondary_emotions":[{"emotion":"curious","intensity":0.3}],"emotional_triggers":[{"claim_id":"c1","emotion":"surprised","reason":"significant_negative_metric"}]},"image":{"url":"https://onjblseywainslkhvkav.supabase.co/storage/v1/object/public/article-images/c72560da-0a80-49e1-927a-cf8cb2387d3a.jpg","alt":"녹색 데이터센터 구조물과 네트워크 연결망을 형상화한 AI 인프라 재설계 개념 이미지","caption":null,"source":"og_image","alt_status":"auto"}},"provenance":{"source_chain":["expert_column","academic_research"],"original_source_url":null,"related_articles":[]},"temporal":{"freshness":"recent","next_update_expected":null},"access":{"license":"neupai_standard","attribution_required":true,"structured_data":"free","full_text_available":true,"full_text_access":null}}