The IIS Chip Gallery
Mothra (2022)
Additional pictures below, click to see larger versions
by
| Application | Communication |
| Technology | 22 |
| Manufacturer | GF |
| Type | Research Project |
| Package | QFN56 |
| Dimensions | 2500μm x 2000μm |
| Gates | 12 MGE |
| Voltage | 0.8 V |
| Power | 787 mW @887 MHz |
| Clock | 887 MHz |
Mothra integrates two designs: One is a novel pre-processing engine and the other a novel energy-efficient, nonlinear data detector for massive MIMO.
Massive multiuser (MU) MIMO enables concurrent transmission of multiple users to a multi-antenna basestation (BS). To detect the users' data using linear equalization, the BS must perform preprocessing, which requires the inversion of a matrix whose dimension equals the number of user data streams. Explicit inversion of large matrices is notoriously difficult to implement due to high complexity, stringent data dependencies that lead to high latency, and high numerical precision requirements. The first design in Mothra is a novel preprocessing architecture based on the block-LDL matrix factorization, which improves parallelism and hence reduces latency. This is the first fabricated preprocessing engine for massive MU-MIMO with 64 BS antennas and 16 single-antenna users.
The second design in Mothra a Gram-domain block coordinate descent algorithm with a deep-unfolding-optimized posterior mean estimate denoiser. It supports 16 UEs transmitting QPSK to 256-QAM symbols to a 128-antenna BS. Measurement results demonstrate a peak throughput of 7.1 Gbps at 367mW with a core area of only 0.97 mm2, demonstrating the detectr's best-in-class throughput and area efficiency.
Following Garamon and Gezora, Mothra is the third in a series of chips depicting Kaiju monsters. Mothra, the name of this ASIC, is a gigantic monster moth, first appeared in "Godzilla: King of the Monsters" (2019)
The chip was manufactured in 22FDX with generous support from GlobalFoundries' University MPW Program.