Die-to-Die Solution Opening the New Era of Flagship SoC

Article By : GUC

Global Unichip Corp. (GUC) has demonstrated the silicon-proven GLink (GUC multi-die interLink) interface using TSMC 7nm process and TSMC InFO_oS advanced packaging technology for AI, HPC and Networking applications to do multi-die integration for system scaling...

Global Unichip Corp. (GUC), the Advanced ASIC Leader, disclosed today that it has successfully demonstrated the silicon-proven GLink (GUC multi-die interLink) interface using TSMC 7nm process and TSMC InFO_oS advanced packaging technology for AI, HPC and Networking applications to do multi-die integration for system scaling.

GLink over InFO_oS is adopted due to InFO_oS cost efficiency for modular, scalable and high-yield multi-die ASICs. GLink over CoWoS is adopted by customers using multi-die ASICs with HBM memories. GLink’s low area/power overhead for high throughput interconnect enables efficient multi-die InFO_oS and CoWoS solutions up to 2500mm2.

Error-free communication between dies with full duplex 0.7 Tbps traffic per 1 mm of beachfront, consuming just 0.25 pJ/bit (0.25W per 1 Tbps of full duplex traffic) was demonstrated. Testing results are fully correlated with pre-silicon simulations in all process-voltage-temperature corners. Early adopting customers are provided with detailed testing reports.

GLink’s power consumption is 6 to 10 times lower than alternative solution using ultra-short reach SerDes-based communication through package substrate. For every 10 Tbps of full duplex traffic it consumes 15 to 20 W less power than alternative SerDes-based interface. GLink IP occupies twice less silicon area and it supports both InFO_oS and CoWoS die integration platforms.

Next generation GLink IP supporting 1.3 Tbps error-free full duplex traffic per 1 mm of beachfront with the same 0.25 pJ/bit power consumption is already available using TSMC 5nm process. Following generation of GLink supporting 2.7 Tbps/mm error-free full duplex traffic with the same 0.25 pJ/bit power consumption using TSMC 5nm and 3nm process will be available during 2021. Such low power/area and traffic per beachfront efficiency makes GLink IPs perfect for AI, HPC and Networking applications.

SerDes-based chips consume constant power according to absolute worst case traffic scenario. GLink-based chips consume power according to actual traffic and data pattern. SerDes-based interfaces randomize data and consume constant power according to worst case data pattern. They always consume the same amount of Watts even when traffic is reduced and data is not random. GLink parallel bus doesn’t randomize data, it consumes power proportionally to actual data toggle rate and even further reduces the toggle rate using DBI. It allows our customers to consume 15 to 20 times less power in practical use cases than if they used SerDes-based links.

Subscribe to Newsletter

Leave a comment