Home | english  | Impressum | Datenschutz | KIT

On the Basis of Brain: Neural Network-Inspired Changes in General Purpose Chips

On the Basis of Brain: Neural Network-Inspired Changes in General Purpose Chips
Datum:

Dec 12th, 2019

Ort:

building 20.30 /

room 0.016

Autoren:

University of Sussex/Brighton

Referent:

Simone Vanucchini

Abstract:

In this paper, we disentangle the changes that the rise of Artificial Intelligence Technologies (AITs) is inducing in the semiconductor industry. The prevailing architectures at the core of the established ‘intensive’ technological trajectory of chip production are currently challenged by increasing data workloads and the rising difficulty to improve product performance over a growing set of computational tasks. More precisely, the sequential model of computation constituting the dominant chip architectures encounters serious competition from concurrent class of computation models. In particular, this is triggered by the use of Artificial Neural Networks (ANNs), a type of program with inherent parallel nature. In turn, this challenge open room for competition and pushes for an adequate response from hardware producers in the form of exploration of new chip architectures and designs.

 

In order to understand this process, we derive a techno-economic trilemma that serves simultaneously as focal points of demand interest and as directions of product improvement for the semiconductor industry players. The trilemma includes (i) computing power, (ii) heterogeneity of computation, and (iii) energy efficiency, and it guides the decisions of producers regarding their production strategy: whether to fork the technological trajectory into sub--trajectories, each built around an own dominant design, or to introduce architectural innovations, making a long term bid on a new dominant design for chips. In fact, the response from chip producers is already visible. Digital giants acquire startups and found departments working on AIT-based software solutions and novel hardware architectures to run them. Nvidia’s Tesla line of GPU--based products, array processors such as Google’s 3rd Generation of TPU, Intel Nervana NNP and MIT-Amazon joint venture chip Eyeriss 2.0, neuromorphic processors like TrueNorth from IBM and Loihi from Intel, and finally radically new optical chip from Lightelligence are among the most recent examples.

 

Pooling together the trilemma and an analysis of the economic forces at work, we construct a simple model showing the mechanism shaping demand in the semiconductor industry. To conclude, we derive two possible scenarios for chip evolution: (i) the emergence of new dominant design in form of ‘platform chip’ comprising heterogeneous cores; (ii) the fragmentation of the semiconductor industry into submarkets with corresponding ASICs. The convergence toward one of the proposed scenarios is conditional on (i) technological progress along the trilemma's edges, (ii) advances in the software domain and its compatibility with hardware, (iii) amount of tasks successfully addressed by this software, (iv) market structure and dynamics.