Architecture

Breakout 1

​

Questions to Ponder

• What do you think are the current major breakthroughs that didn’t get mentioned during those talks?
• What do you think has to happen in these areas for quantum computing to become viable in N years (10 or 20)?
• How can computer scientists play a role?

What do you think are the current major breakthroughs that didn’t get mentioned during those talks?

• New and improved simulation capabilities
• Architects are used to working with simulation
• How do we tractably simulate noise in circuits?
• Recent work enables various approximate simulation strategies
• Potential for approximate computing applications to quantum
• There are type systems that track resources
• We could track energy, power, error
• Any advantage for mixed digital/analog algorithms?
• Does it help in the algorithm to switch what types of errors are more likely?
• The end-to-end compilation flow has to happen to understand resource usage as most algorithms are just subroutines
• Need for heuristics to merge subroutines on boundaries
• In distributed computing the programmer has the ability to schedule some of the execution
• For near-term device do we need to think in the circuit model? Why not just optimize Hamiltonians
• There are some examples such as global CNOTs that can be done in one shot
• In a mixed mode, one could profile the application and optimize at hot spots - there are trade offs if parameters are not static
• Globally-optimal circuit optimizers for small systems
• Potential use for proof assistants

​

What do you think has to happen in these areas for quantum computing to become viable in N years (10 or 20)?

• For the scale of machines in 10-20 years (10,000 - 100,000 qubits say) we need to define abstractions layers
• This will not be as optimal, but we have resources at disposal
• It may be too soon for an abstracted ISA, there is a need for abstraction as compilation from high-level to see of qubits/control pulses is not feasible at scale
• There is a current path from Scaffold -> open QASM -> Qiskit that goes from high-level to qubits. Only designed for small scale
• A common optimization framework at different layers of software
• This requires a scalable optimization strategy
• A rich interface allows for optimizations that may allow for some breaking of abstraction
• Verified compilers
• Long distance communication (e.g. optical transduction) would give you the ability to build modular architectures
• Hybrid architectures
• Architects interacting with hardware people

​

How can computer scientists play a role?

• Abstraction layers
• SIMD architecture possibilities
• Can reduce resources for both FT and non-FT architectures
• 11 - 3 instructions for physical Toffoli gate
• It works well for small circuits, what about larger algorithms?

​