I always fall in love with an open door
Galen Reeves, Jie Liu, Suman Nath, Feng Zhao: Managing Massive Time Series Streams with Multi-Scale Compressed Trickles.Vijay Janapa Reddi, Benjamin Lee, Trishul Chilimbi, Kushagra Vaid: Web Search Using Small Cores: Quantifying the Price of Efficiency.
A 32-bit ARM can drive one DIMM. You could fit 32 or more Cortex-A9s on a chip, but the result would be completely unbalanced with only 256 MB RAM per core. Perhaps the opposite design point is better: only two cores, 4-8 MB of cache, and a single-channel memory controller. The resulting chip is very cheap with high yield because it is mostly cache. The power is low enough that you could put it on a DIMM like an AMB. You could call it BlueGene/ARM. :-)
