Invest1 publisher3 min readPublished Updated
Shinko's 22-layer glass substrate turns the AI packaging race into a heat problem
Japan's Shinko says a polymer edge coat stopped glass cracking under thermal load. Samsung Electro-Mechanics is chasing mass production through a $346 million joint venture with Dongwoo Fine-Chem.
The Investor · Invest desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Japan's Shinko Electric Industries has built a 22-layer glass substrate structure by stacking 11 copper wiring layers on each side of a glass substrate, according to industry sources cited on the 15th.
- The Shinko achievement solves the so-called 'seware' problem, in which cracks or delamination occur inside the glass.
- Shinko succeeded in suppressing seware even after thermal loads were applied.
- Who can first mass-produce a glass substrate with high thermal reliability is emerging as the decisive battleground for capturing the market, amid an intensifying technology race between South Korea and Japan over glass substrates for AI chips.
- AI chips are evolving toward integrating multiple processing chips and memory chips into a single package, so package sizes are growing and the limits of conventional organic (polymer) substrates, which warp as their area widens, are becoming more apparent.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
Shinko Electric Industries has built a 22-layer glass substrate by stacking 11 copper wiring layers on each side of a glass core, and says it has suppressed "seware" - the internal cracking and delamination that glass suffers under heat - even after thermal loading [1][2][3]. That result moves the Korea-Japan glass substrate contest off the question of who can pattern glass and onto a narrower one: who can mass-produce glass that survives a reflow oven [4].
The reason anyone cares is package size. AI chips are being built by integrating multiple logic and memory dies into a single package, and conventional organic substrates warp as their area widens [5]. Glass is rigid and barely deforms thermally, which is why substrate makers are treating it as the next platform rather than a variant [6]. Shinko's fix was materials, not lithography: a polymer-based protective material applied to the substrate edges to disperse the stress that concentrates in specific areas when heat is applied [7].
Samsung Electro-Mechanics is coming at the same problem through structure. Last month it signed a definitive agreement to form GlaSSEM with Dongwoo Fine-Chem, the Korean subsidiary of Sumitomo Chemical, capitalized at 482.1 billion won, about $346 million, split 66% to Samsung Electro-Mechanics and 34% to Dongwoo [8]. The division of labour is explicit: Dongwoo brings large-area glass processing and production, Samsung Electro-Mechanics brings high-layer-count, high-density substrate design and mass-production know-how, and the venture owns glass core development through to sales [9]. On its second-quarter call the company said it is co-developing large-area, high-layer-count glass substrates for data centres with global big tech firms and running technology approval and sample evaluation with some customers [10]. It is targeting mass production two years out [11].
The industry read, based on disclosed results, is that Japan is a step ahead [12]. Shinko has both through-glass-via process technology and the multilayer insulation and copper stacking that sits on top of it [13]. Dai Nippon Printing built a TGV pilot line in Saitama Prefecture last December, has shipped samples since early this year, and plans full mass production in 2028 [14]. Samsung Electro-Mechanics' two-year target lands in the same year [1].
The revenue pool, for now, is small. Persistence Market Research projects the global glass core substrate market at $128.1 million this year rising to $1.036 billion in 2033, a 34.8% compound growth rate [15] - roughly 8.1 times over seven years [2]. GlaSSEM's capitalization alone is about a third of that projected 2033 market [3]. Nobody is committing that capital for substrate margins; they are committing it to hold position in AI packaging, next to an FC-BGA business Samsung Electro-Mechanics already mass-produces at high margin [16].
Two things to watch. First, whether Shinko's 22-layer result survives customer qualification rather than internal test, since thermal reliability claims tend to degrade once a substrate is populated with dies and reflowed at scale. Second, yield on large-area glass at GlaSSEM, which is the variable Dongwoo was brought in for and the one that decides whether the 2028 dates from Tokyo and Suwon mean the same thing [9][14][11].