SAINT-SULPICE, SWITZERLAND — June 3,2026 — Kandou AI, a leading fabless semiconductorcompany specializing in high-speed, energy-efficient connectivity and systemsolutions for the AI era, today announced the successful silicon demonstrationof its industry-leading Tigerwing™ Chip-to-Chip (C2C) SerDes interface onTSMC’s 4nm technology node. The demonstration took place at the ProductInnovation Demos platform during the 2026 TSMC Europe Technology Symposium’s“Innovate with TSMC!” showcase, May 28th. This breakthrough marks a majormilestone in Kandou AI’s mission to solve the critical data movementbottlenecks hampering modern AI infrastructure.

Solving the AI Memory and Interconnect Wall

As artificial intelligence models grow exponentially, the semiconductor industry faces a fundamental roadblock: GPU sand CPUs must connect to massively expanding memory footprints, turning memory and interconnect bandwidth into severe systemic bottlenecks. The performance of modern AI clusters is no longer limited by raw compute power, but by the speed and efficiency of the data pipelines connecting that compute infrastructure to memory. Traditional networking and interconnect technologies, designed for a pre-AI world, simply cannot keep pace.

Kandou AI’s silicon-proven, revolutionary interconnect technologies are purpose-built to eliminate these exact bottlenecks. By operating at the intersection of information theory and semiconductor design, Kandou AI’s patented signaling and SerDes technologies fundamentally reimagine how data moves across copper infrastructure—delivering breakthrough improvements in bandwidth, reach, and energy efficiency. Unlocking this level of performance from copper allows for the construction of scalableAI systems at a fraction of traditional costs and power consumption, accelerating the global democratization of AI infrastructure.

Turning the Entire PCB into a Large MCM

At the TSMC Symposium, Kandou AI showcased its Tigerwing™ C2C, a groundbreaking SerDes link that delivers long-distance chip-to-chip connectivity across standard PCBs at performance levels previously only achievable via Die-to-Die (D2D) links inside a single package.

This advancement is a game-changer for system design, effectively transforming an entire PCB into a massive Multi-ChipModule (MCM). Chiplet design spaces are no longer constrained to a single package; the Tigerwing™  link is engineered to connect two separate chips mounted on two different package sacross a PCB distance exceeding 200 mm (see Figure 1).

This architectural shift enables dramatically scaled, lower-cost memory connectivity for advanced AI systems.The live demonstration successfully showcased copper links running at 260Gbps over a 200 mm reach, achieving an industry-leading power efficiency of< 1pJ/bit at a Bit Error Rate (BER) of < 1E-19 (as shown in Figure 2).

‍

‍

“This demonstration is a major milestone for Kandou AI as it validates our new, industry-leading high-speed chip-to-chip interconnect technology. With Tigerwing™  technology at the core, we are building transformative AI systems which solve key connectivity and capacity bottlenecks for our AI datacenter customers” said Srujan Linga, Co-Founder and CEO of Kandou AI.  

‍

Link to video presentation

About Kandou AI

Kandou AI is a fabless semiconductor design company dedicated to advancing high-speed, longer reach, low power connectivity solutions. Leveraging its differentiated interconnect IP and system design capabilities, Kandou AI builds high-performance connectivity products for its customers across datacenter infrastructure, AI and consumer electronics with over 28mm silicon units shipped using its technology. Kandou AI is headquartered in Saint-Sulpice, Switzerland, with offices worldwide.

‍