{"@context":"https://neupai.io/schema/v0.2","@type":"StructuredNewsArticle","identity":{"article_id":"skhynix_20261020_ai-ecosystem-memory-paradigm","canonical_url":"https://news.skhynix.co.kr/ai-ecosystem-series-ep4/","ai_url":null,"publisher":{"name":"SK하이닉스 뉴스룸","domain":"news.skhynix.co.kr","type":"online"},"author":"unknown","published_at":"2026-10-20T23:59:56.000Z","updated_at":null,"language":"en","article_type":"feature","originality":"self_produced"},"content":{"headline":"[AI Ecosystem] The Shift in Semiconductor Paradigm: Toward Memory-Centric Design","summary":"The focus of AI chip design is shifting from computation to data supply and memory, with HBM and PIM technologies emerging as the core structures supporting this transition.","topics":["semiconductors","AI","memory","HBM","packaging"],"geography":["KR","US"],"entities":[{"name":"SK Hynix","canonical_id":"corp:kr:sk-hynix","type":"company","role_in_article":"primary_subject","metadata":{"ticker":"000660.KS","parent":null}},{"name":"Raja Koduri","canonical_id":"person:us:raja-koduri","type":"person","role_in_article":"quoted","metadata":{}},{"name":"OXMIQ Labs","canonical_id":"corp:us:oxmiq-labs","type":"company","role_in_article":"mentioned","metadata":{"ticker":null,"parent":null}},{"name":"AMD","canonical_id":"corp:us:amd","type":"company","role_in_article":"mentioned","metadata":{"ticker":"AMD","parent":null}},{"name":"Intel","canonical_id":"corp:us:intel","type":"company","role_in_article":"mentioned","metadata":{"ticker":"INTC","parent":null}},{"name":"Apple","canonical_id":"corp:us:apple","type":"company","role_in_article":"mentioned","metadata":{"ticker":"AAPL","parent":null}},{"name":"Google","canonical_id":"corp:us:google","type":"company","role_in_article":"mentioned","metadata":{"ticker":"GOOGL","parent":null}}],"claims":[{"id":"c1","statement":"Gordon Moore proposed Moore's Law in 1965","as_of":"1965","as_of_explicit":true,"as_of_raw":"1965","source_type":"research_paper","comparison":null,"type":"fact","figures":null,"expiry_hint":null,"insight":null},{"id":"c2","statement":"A petaflop refers to floating-point computing performance of about 1 quadrillion (10^15) operations per second","as_of":"2026-10","as_of_explicit":false,"as_of_raw":"October 2026","source_type":"industry_estimate","comparison":null,"type":"fact","figures":{"value":1000000000000000,"unit":"operations_per_second","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c3","statement":"Fetching data from off-chip DRAM can incur energy costs hundreds to thousands of times greater than from on-chip SRAM","as_of":"2026-10","as_of_explicit":false,"as_of_raw":"October 2026","source_type":"industry_estimate","comparison":"on_chip_sram","type":"fact","figures":{"value":1000,"unit":"ratio","approximate":true,"converted":null},"expiry_hint":null,"insight":null},{"id":"c4","statement":"The Transformer was introduced by Google in 2017","as_of":"2017","as_of_explicit":true,"as_of_raw":"2017","source_type":"company_disclosure","comparison":null,"type":"fact","figures":null,"expiry_hint":null,"insight":null},{"id":"c5","statement":"The author participated in developing an HBM-equipped GPU at AMD in 2015","as_of":"2015","as_of_explicit":true,"as_of_raw":"2015","source_type":"company_disclosure","comparison":null,"type":"fact","figures":null,"expiry_hint":null,"insight":null},{"id":"c6","statement":"Intel's development of Ponte Vecchio integrated 47 chiplets and 4 HBM2e stacks into a single package","as_of":"2026-10","as_of_explicit":false,"as_of_raw":"October 2026","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":47,"unit":"chiplets","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c7","statement":"Using PIM can reduce energy consumption by 10 to 100 times depending on the workload","as_of":"2026","as_of_explicit":false,"as_of_raw":"2026","source_type":"research_paper","comparison":"conventional_computing","type":"estimate","figures":{"value":100,"unit":"ratio","approximate":true,"converted":null},"expiry_hint":null,"insight":{"analyst_name":"Raja Koduri","analyst_affiliation":"OXMIQ Labs","confidence":"medium","forecast_horizon":"long_term","sentiment":"bullish","reasoning":"PIM can significantly improve efficiency in tasks where the cost of data movement exceeds the cost of computation"}}],"ai_emotional_context":{"valence":0.4,"arousal":0.5,"primary_emotions":[{"emotion":"optimistic","intensity":0.7},{"emotion":"excited","intensity":0.5}],"secondary_emotions":[{"emotion":"curious","intensity":0.4}],"emotional_triggers":[{"claim_id":"c7","emotion":"excited","reason":"major_opportunity"}]},"image":{"url":"https://onjblseywainslkhvkav.supabase.co/storage/v1/object/public/article-images/c5780b47-dd26-4f08-b428-9dbe76ca1722.jpg","alt":"회로기판 위에 두 개의 반도체 칩이 빛나는 데이터 연결로 이어진 모습, 메모리 중심 반도체 패러다임 전환을 상징","caption":null,"source":"og_image","alt_status":"auto"}},"provenance":{"source_chain":["expert_column","SK_hynix_newsroom"],"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}}