Application | Machine-learning |
Technology | 65 |
Manufacturer | UMC |
Type | Semester Thesis |
Package | QFN40 |
Dimensions | 1252μm x 1252μm |
Gates | 600 kGE |
Voltage | 1.2 V |
Power | 30.36 mW @ 1.275V, 159 MHz |
Clock | 159 MHz |
This is an iteration of the Chipmunk design, a multi-chip systolically scalable hardware architecture for Long short-term memory networks (LSTM) recurrent neural networks (RNNs). The accelerators keep all parameters stationary on every die in the array, drastically reducing the I/O communication to only loading new features and sharing partial results with other dies. The main differences to Chipmunk are a modified clocking scheme, a changed pin layout, and additional on-chip logic to facilitate the construction of a demonstrator on the systolic scaling with minimal additional components on the PCB.
When Muntaniala was taped-out, RNNs were state-of-the-art in voice awareness/understanding and speech recognition. On-device computation of RNN on low-power mobile and wearable devices was key to applications such as zero-latency voice-based human-machine interfaces.
The name is the Rumantsch (the fourth official language of Switzerland) word for the Marmot. Gianna comes from Graubuenden, the Canton of Switzerland, where Rumantsch is spoken.