Add reverse-compatibility statement

1 files changed, 41 insertions, 4 deletions

diff --git a/bip-powx-low-energy-pow.mediawiki b/bip-powx-low-energy-pow.mediawiki
index bf72893..0e0070e 100644
--- a/bip-powx-low-energy-pow.mediawiki
+++ b/bip-powx-low-energy-pow.mediawiki
@@ -43,7 +43,7 @@ for billions of new miners to enter the market simply by investing in a
 low-energy photonic miner. Shifting to a high-CAPEX PoW has the added benefit of
 making the hashrate resilient to Bitcoin's price fluctuations - once low-OPEX
 hardware is operating there is no reason to shut it down even if the value of
-mining rewards diminishes. oPoW is backward compatible with GPUs, FPGAs, and
+mining rewards diminishes. oPoW is hardware-compatible with GPUs, FPGAs, and
 ASICs meaning that a transitional period of optical and traditional hardware
 mining in parallel on the network is feasible
 
@@ -69,7 +69,22 @@ Whether on not the Bitcoin community accepts this common criticism as entirely v
 
 New consensus mechanisms have been proposed as a means of securing cryptocurrencies whilst reducing energy cost, such as various forms of Proof of Stake and Proof of Space-Time. While many of these alternative mechanisms offer compelling guarantees, they generally require new security assumptions, which have not been stress-tested by live deployments at any adequate scale. Consequently, we still have relatively little empirical understanding of their safety. Completely changing the Bitcoin paradigm is likely to introduce new unforeseen problems. We believe that the major issues discussed above can be resolved by improving rather than eliminating Bitcoin’s fundamental security layer—Proof of Work. Instead of devising a new consensus architecture to fix these issues, it is sufficient to shift the economics of PoW. The financial cost imposed on miners need not be primarily composed of electricity. The situation can be significantly improved by reducing the operating expense (OPEX)—energy—as a major mining component. Then, by shifting the cost towards capital expense (CAPEX)—mining hardware—the dynamics of the mining ecosystem becomes much less dependent on electricity prices, and much less electricity is consumed as a whole.
 
-Moreover, a reduction in energy consumption automatically leads to geographically distributed mining, as mining becomes profitable even in regions with expensive electricity. Additionally, lower energy consumption will eliminate heating issues experienced by today’s mining operations, which will further decrease operating cost as well as noise associated with fans and cooling systems. All of this means that individuals and smaller entities would be able to enter the mining ecosystem simply for the cost of a miner, without first gaining access to cheap energy or a dedicated, temperature-controlled data center. To a degree, memory-hard PoW schemes like [https://github.com/tromp/cuckoo Cuckoo Cycle], which increase the use of SRAM in lieu of pure computation, push the CAPEX/OPEX ratio in the right direction by occupying ASIC chip area with memory. To maximize the CAPEX to OPEX ratio of the Optical Proof of Work algorithm, we developed [https://assets.pubpub.org/xi9h9rps/01581688887859.pdf ''HeavyHash''] [1]. HeavyHash is a cryptographic construction that takes the place of SHA256 in Hashcash. Our algorithm is compatible with ultra-energy-efficient photonic co-processors that have been developed for machine learning hardware accelerators.
+Moreover, a reduction in energy consumption automatically leads to
+geographically distributed mining, as mining becomes profitable even in regions
+with expensive electricity. Additionally, lower energy consumption will
+eliminate heating issues experienced by today’s mining operations, which will
+further decrease operating cost as well as noise associated with fans and
+cooling systems. All of this means that individuals and smaller entities would
+be able to enter the mining ecosystem simply for the cost of a miner, without
+first gaining access to cheap energy or a dedicated, temperature-controlled data
+center. To a degree, memory-hard PoW schemes like
+[https://github.com/tromp/cuckoo Cuckoo Cycle], which increase the use of SRAM
+in lieu of pure computation, push the CAPEX/OPEX ratio in the right direction by
+occupying ASIC chip area with memory. To maximize the CAPEX to OPEX ratio of the
+Optical Proof of Work algorithm, we developed
+[https://assets.pubpub.org/xi9h9rps/01581688887859.pdf ''HeavyHash''] [1].
+HeavyHash is a cryptographic construction that takes the place of SHA256 in
+Hashcash. Our algorithm is hardware-compatible with ultra-energy-efficient photonic co-processors that have been developed for machine learning hardware accelerators.
 
 HeavyHash uses a proven digital hash (SHA3) packaged with a large amount of MAC (Multiply-and-Accumulate) computation into a Proof of Work puzzle. Although HeavyHash can be computed on any standard digital hardware, it becomes hardware efficient only when a small digital core is combined with a low-power photonic co-processor for performing MAC operations. oPoW mining machines will have a small digital core flip-chipped onto a large, low-power photonic chip. This core will be bottlenecked by the throughput of the digital to analog and analog to digital converters. A prototype of such analogue optical matrix multiplier can be seen in the figure below.
 
@@ -79,7 +94,21 @@ Figure. TOP: Photonic Circuit Diagram, A. Laser input (1550nm, common telecom wa
 
 The ''HeavyHash'' derives its name from the fact that it is bloated or weighted with additional computation. This means that a cost comparable oPoW miner will have a much lower nominal hashrate compared to a Bitcoin ASIC (HeavyHashes/second vs. SHA256 Hashes/second in equivalent ASIC). We provide the cryptographic security argument of the HeavyHash function in Section 3 in [https://assets.pubpub.org/xi9h9rps/01581688887859.pdf Towards Optical Proof of Work] [1]. In the article, we also provide a game-theoretic security argument for CAPEX-heavy PoW. For additional information, we recommend reading [https://uncommoncore.co/wp-content/uploads/2019/10/A-model-for-Bitcoins-security-and-the-declining-block-subsidy-v1.02.pdf this article].
 
-While traditional digital hardware relies on electrical currents, optical computing uses light as the basis for some of or all of its operations. Building on the development and commercialization of silicon photonic chips for telecom and datacom applications, modern photonic co-processors are silicon chips made using well-established and highly scalable silicon CMOS processes. However, unlike cutting edge electronics which require ever-smaller features (e.g. 5 nm), fabricated by exponentially more complex and expensive machinery, silicon photonics uses old fabrication nodes (90 nm). Due to the large de Broglie wavelength of photons, as compared to electrons, there is no benefit to using the small feature sizes. The result is that access to silicon photonic wafer fabrication is readily available, in contrast to the notoriously difficult process of accessing advanced nodes. Moreover, the overall cost of entry is lower as lithography masks for silicon photonics processes are an order of magnitude cheaper ($500k vs. $5M). Examples of companies developing optical processors for AI, which will be compatible with oPoW include [https://lightmatter.co/ Lightmatter], [https://www.lightelligence.ai/ Lightelligence], [https://luminous.co/ Luminous], [https://www.intel.com/content/www/us/en/architecture-and-technology/silicon-photonics/silicon-photonics-overview.html Intel], and other more recent entrants.
+While traditional digital hardware relies on electrical currents, optical
+computing uses light as the basis for some of or all of its operations. Building
+on the development and commercialization of silicon photonic chips for telecom
+and datacom applications, modern photonic co-processors are silicon chips made
+using well-established and highly scalable silicon CMOS processes. However,
+unlike cutting edge electronics which require ever-smaller features (e.g. 5 nm),
+fabricated by exponentially more complex and expensive machinery, silicon
+photonics uses old fabrication nodes (90 nm). Due to the large de Broglie
+wavelength of photons, as compared to electrons, there is no benefit to using
+the small feature sizes. The result is that access to silicon photonic wafer
+fabrication is readily available, in contrast to the notoriously difficult
+process of accessing advanced nodes. Moreover, the overall cost of entry is
+lower as lithography masks for silicon photonics processes are an order of
+magnitude cheaper ($500k vs. $5M). Examples of companies developing optical
+processors for AI, which will be hardware-compatible with oPoW include [https://lightmatter.co/ Lightmatter], [https://www.lightelligence.ai/ Lightelligence], [https://luminous.co/ Luminous], [https://www.intel.com/content/www/us/en/architecture-and-technology/silicon-photonics/silicon-photonics-overview.html Intel], and other more recent entrants.
 
 == Specification ==
 
@@ -91,7 +120,10 @@ The HeavyHash is performed in three stages:
 # Matrix-vector multiplication
 # Keccak of the result xorred with the hashed input
 
-Note that the most efficiently matrix-vector multiplication is performed on a photonic miner. However, this linear algebra operation can be performed on any conventional computing hardware (CPU, GPU, etc.), therefore making the HeavyHash compatible with any digital device.
+Note that the most efficiently matrix-vector multiplication is performed on a
+photonic miner. However, this linear algebra operation can be performed on any
+conventional computing hardware (CPU, GPU, etc.), therefore making the HeavyHash
+hardware-compatible with any digital device.
 
 The algorithm’s pseudo-code:
 
@@ -239,6 +271,11 @@ Although it is out of the scope of this proposal, the authors strongly recommend
 
 A hard fork is not necessarily required for the Bitcoin network to test and eventually implement oPoW. It’s possible to add oPoW as a dual PoW to Bitcoin as a soft fork. Tuning the parameters to ensure that, for example, 99.9% of the security budget would be earned by miners via the SHA256 Hashcash PoW and 0.1% via oPoW would create sufficient incentive for oPoW to be stress-tested and to incentivize the manufacture of dedicated oPoW miners. If this test is successful, the parameters can be tuned continuously over time, e.g. oPoW share doubling at every halving, such that oPoW accounts for some target percentage (up to 100% in a complete SHA256 phase-out).
 
+
+==== Reverse compatibility ====
+
+In general, we do not claim reverse compatibility, meaning it may not be possible to fully implement oPoW and maintain compatibility with older versions of the codebase that only perform SHA256 mining.
+
 === ASICBOOST ===
 
 Any new PoW algorithm carries the risk of hardware developers discovering and patenting an architecture with a significant speedup, as happened in the case of ASICBOOST for SHA256. HeavyHash is comprised of an SHA hash and 4-bit linear matrix-vector operations. The intent is for the matrix-vector multiplications to account for the majority of the work involved in computing a single HeavyHash operation. As we show in the Minimum Effective Hardness section of Towards Optical Proof of Work[1], there is no workaround to performing the matrix operations when computing HeavyHash, and since the SHA hashes are negligible, a true ASICBOOST-type speed up would require a speed up in linear matrix processing. Since matrix-vector multiplication is at the heart of neural networks and many other common computational workloads, it has been optimized very heavily and is generally very well understood. The acceleration of matrix-vector multiplication hardware (e.g. photonic coprocessors, memristors, etc.) is a very general problem and there are dozens of companies working on it, making it very unlikely for a single party to corner the market.