Who We Are

POCONS USA designs and manufactures board-level EMI shields, shield clips, spring contacts, and custom metal stampings. POCONS USA is headquartered in San Diego, with products manufactured in Korea (IATF 16949) and high-volume stamping and plating in Vietnam. That manufacturing footprint gives us a direct read on the Korean component supply chain weeks before it reaches a US distributor's portal. This newsletter exists to pass that signal forward.

Supply Chain Alert: Naphtha Starvation Reaches the Cracker

When a steam cracker goes cold, it does not "pause." It freezes coke onto the radiant coils, and the restart is measured in months, not days.

South Korea's petrochemical sector is running on fumes. Roughly 77% of Korean naphtha feedstock historically transits the Strait of Hormuz, and the closure has forced LG Chem and Yeochun NCC (YNCC) to idle crackers with inventory buffers measured in single-digit days. This is not abstract. Naphtha cracking yields ethylene and propylene — the precursors for the epoxy molding compounds, BT/ABF substrate resins, and the polymer dielectrics in conductive-polymer capacitors.

Leading Korean memory makers represent ~70% of global DRAM and ~80% of HBM output. DRAM contract pricing has already climbed 95% quarter-over-quarter, and HBM3E/HBM4 capacity is fully committed through 2026. A feedstock interruption that propagates into substrate and EMC supply does not just raise prices — it caps the number of finished memory stacks that can be assembled at all.

⚠️Procurement action

Any BOM line with a Korean-fabricated organic substrate (FC-BGA, FC-CSP) or epoxy molding compound should be re-confirmed for allocation status this week. The lead-time signal will arrive before the price signal.

Price Watch: Metals and the MLCC Rotation

Base-metal prices have decoupled from soft demand and are tracking energy and tariff risk instead.

Tin matters disproportionately to anyone reflowing a board: SAC305 solder is 96.5% tin. A move from $47,400 to $51,613/t is a ~9% input-cost shock to every solder joint, and tin is also a plating constituent on EMI shield frames.

This issue's component rotation focuses on MLCCs:

Standard 0402/0603 — Tightening. Murata and the leading Korean MLCC makers have pushed double-digit increases through; Yageo and Walsin are following on the same cadence as the 2018 upcycle.
High-capacity / automotive — Allocation. A single AI accelerator board consumes 10,000–20,000 MLCCs for power-rail decoupling. That demand competes head-to-head with automotive lines for the same X7R/X6S capacity.
High-voltage (≥1 kV) — Tight. 800 V EV traction platforms drive 2 kV-rated X7R demand; new families are selling through before they reach broad distribution.

Korean Manufacturer Spotlight: Leading Korean MLCC Makers

The top-tier Korean MLCC makers rank among the world's largest passive suppliers, with fabrication across multiple Korean sites. Their product lines span MLCCs, power inductors, FC-BGA substrates, and camera modules.

Why US engineers should care: the leading Korean MLCC makers are the price bellwether. When they move MLCC pricing, Murata, Yageo, and Walsin follow within weeks — so their announcements are a leading indicator for your whole passive BOM. High-voltage 2 kV X7R families from these suppliers were built specifically for 800 V SiC inverter snubber and DC-link decoupling, and their FC-BGA substrates compete with Ibiden for AI-server packaging. Analyst projections suggest spare capacity at top-tier Korean suppliers exhausts by H2 2026 — a simultaneous MLCC and FC-BGA upcycle not seen since the 5G buildout.

POCONS connection: POCONS USA's products are manufactured in Korea, giving us direct visibility into the Korean component supply chain. When top-tier supplier capacity tightens, our EMI shield customers feel it downstream — because the same boards needing 15,000 MLCCs also need board-level shielding over the RF and power sections.

Korean Supply Chain Intelligence

Korea posted a record ~$30B semiconductor export month in March on AI-memory demand — but trade officials are openly warning that Q2 output is at risk if petrochemical feedstock is not stabilized.
Korea imported Russian naphtha for the first time post-Hormuz — 27,000 tonnes under a temporary US sanctions waiver. One cargo sustains minimal cracker operation; it does not refill a two-week buffer that gates every Korean-sourced resin, substrate, and molding compound in your BOM.

Design Corner: MLCC Placement Is an Inductance Problem

The decoupling capacitor you specified is not the capacitor the IC sees. The IC sees the capacitor plus the loop inductance of the connection. PCB trace inductance runs ≈ 0.8–1.0 nH/mm, and via inductance adds ≈ 0.3–0.5 nH per via.

A 100 nF 0402 (ESL ≈ 0.4 nH intrinsic) on a 3 mm trace with two vias sees a total loop inductance of roughly:

L_total ≈ 0.4 + (3 × 0.9) + (2 × 0.4) ≈ 3.9 nH

Self-resonant frequency:

f_SRF = 1 / (2π√(L·C)) = 1 / (2π√(3.9 nH × 100 nF)) ≈ 8 MHz

Above f_SRF the capacitor is inductive — it stops decoupling and starts forming a resonant tank with the PDN. Place high-value decoupling within 1 mm of the power-pin via, return on the immediately adjacent plane to minimize loop area, and use a dual-cap strategy: small cap at the pin for high-frequency, bulk cap displaced for charge reservoir.

💡EMI tie-in

A PDN resonance is a radiated-emissions source. The same 80–200 MHz tank you create with sloppy decoupling is exactly the band where board-level shields earn their keep — but a shield treats the symptom. Fix the loop inductance first; shield what remains.

One Thing

A typical AI server board uses 10,000–20,000 MLCCs. A smartphone uses 1,000. The MLCC industry just decided which customer it works for — and the price you pay reflects that decision, not your volume.

— Mike Kwak, POCONS USA